paper_id string | arxiv_id string | title string | markdown dict | reviews list | scores dict | metadata dict | meta_review dict | decision dict |
|---|---|---|---|---|---|---|---|---|
H5FUVj0vMd | 2406.18629v1 | Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs | {
"content": "## Abstract\n\nAbstract Mathematical reasoning presents a significant challenge for Large Language Models (LLMs) due to the extensive and precise chain of reasoning required for accuracy. Ensuring the correctness of each reasoning step is critical. To address this, we aim to enhance the robustness and f... | [
{
"id": "smQMg2sDXc",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes to dpo train the model per each reasoning step. To do so, the authors prepare questions and sampling LLMs to generate reasoning steps, and ann... | {
"rating": "1;3;5;5",
"rating_avg": 3.5,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.489778"
} | {
"id": "o2VQh7Y6Mk",
"metareview": "This paper presents Step-DPO, which is a variant of DPO that use intermediate reasoning steps as the preference feedback to the model.\n\nThe paper is clear, results show that Step-DPO improves performance, and its ablation analyses on DPO vs Step-DPO and in vs out of distributi... | {
"decision": "Reject"
} |
H6DpBnPCyH | 2409.18152v1 | Reinforcement Learning for Finite Space Mean-Field Type Games | {
"content": "## Abstract\n\nAbstract Mean field type games (MFTGs) describe Nash equilibria between large coalitions: each coalition consists of a continuum of cooperative agents who maximize the average reward of their coalition while interacting non-cooperatively with a finite number of other coalitions. Although ... | [
{
"id": "SOPWFhhyv9",
"initial_rating": 3,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper proposes a mean field type game (MFTG) with a finite number of noncooperative coalitions in which each coalition consists of a continuum of cooperative... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "1;3;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.490604"
} | {
"id": "WxbBBb2NM0",
"metareview": "The paper explores reinforcement learning methods for mean-field type games (MFTGs), a setting where cooperative agents within coalitions compete non-cooperatively against other coalitions. The authors propose a Nash Q-learning algorithm with theoretical guarantees and a deep RL... | {
"decision": "Reject"
} |
H6UMc5VS70 | 2410.02832v1 | FlipAttack: Jailbreak LLMs via Flipping | {
"content": "## Abstract\n\nAbstract This paper proposes a simple yet effective jailbreak attack named FlipAttack against black-box LLMs. First, from the autoregressive nature, we reveal that LLMs tend to understand the text from left to right and find that they struggle to comprehend the text when noise is added to... | [
{
"id": "RjECO2EGPT",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 4,
"summary": "This paper proposes a simple yet effective jailbreak attack named FlipAttack against black-box LLMs. First, from the autoregressive nature, the authors reveal tha... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "1;1;2;3",
"soundness_avg": 1.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;4;4;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.492062"
} | {
"id": "6Wf5oxP7Ou",
"metareview": "The paper proposes a new jailbreak by flipping the order of text, words and characters in harmful prompts.\nThe attack is shown to be quite effective, but there is maybe limited scientific value in such attacks as the design is primarily heuristic, and there is little analysis d... | {
"decision": "Reject"
} |
H8hO3T3DYe | 2406.07475v1 | Partially Observed Trajectory Inference using Optimal Transport and a Dynamics Prior | {
"content": "## Abstract\n\nAbstract Trajectory inference seeks to recover the temporal dynamics of a population from snapshots of its (uncoupled) temporal marginals, i.e. where observed particles are not tracked over time. Lavenant et al. [ 33 ] addressed this challenging problem under a stochastic differential equ... | [
{
"id": "GfR5I3mrWR",
"initial_rating": 3,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 1,
"summary": "This paper proposes a latent variable model for trajectory inference, i.e., inferring a time-indexed collection of particles. We assume a latent state vector $(X_... | {
"rating": "3;3;8",
"rating_avg": 4.666666666666667,
"confidence": "3;2;2",
"confidence_avg": 2.3333333333333335,
"soundness": "2;3;4",
"soundness_avg": 3,
"contribution": "2;2;4",
"contribution_avg": 2.6666666666666665,
"presentation": "3;1;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.493234"
} | {
"id": "PtJmEfZaCQ",
"metareview": "This work extends trajectory inference to latent state-space models using partially observed data and specified dynamics (e.g., constant velocity/acceleration). It introduces the PO-MFL algorithm with theoretical guarantees, building on prior stochastic differential equation fra... | {
"decision": "Accept (Poster)"
} |
H9UnNgdq0g | 2409.15477v1 | MediConfusion: Can you trust your AI radiologist? Probing the reliability of multimodal medical foundation models | {
"content": "## Abstract\n\nAbstract Multimodal Large Language Models (MLLMs) have tremendous potential to improve the accuracy, availability, and cost-effectiveness of healthcare by providing automated solutions or serving as aids to medical professionals. Despite promising first steps in developing medical MLLMs i... | [
{
"id": "wJQ0Zl6PTN",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces MediConfusion, a benchmark specifically designed to assess and expose failure modes in multimodal large language models (MLLMs) used in medic... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;4;3;4",
"contribution_avg": 3.25,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.494400"
} | {
"id": "dJmJpZvlH2",
"metareview": "This is a datasets and benchmarks paper. The authors introduced MediConfusion, a challenging medical Visual Question Answering (VQA) benchmark dataset that probes the failure modes of medical MLLMs from a vision perspective. The benchmark dataset is designed to evaluate the abil... | {
"decision": "Accept (Poster)"
} |
HAD6iZxKuh | 2406.08337v1 | WMAdapter: Adding WaterMark Control to Latent Diffusion Models | {
"content": "## Abstract\n\nAbstract Watermarking is crucial for protecting the copyright of AI-generated images. We propose WMAdapter, a diffusion model watermark plugin that takes user-specified watermark information and allows for seamless watermark imprinting during the diffusion generation process. WMAdapter is... | [
{
"id": "hPhJGJgUCr",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper presents WMAdapter, a plug-in solution for diffusion-based watermarking. The authors propose training a contextual adapter and introduce a hybrid fine-... | {
"rating": "3;5;5;6;6",
"rating_avg": 5,
"confidence": "5;5;5;4;4",
"confidence_avg": 4.6,
"soundness": "2;2;2;2;3",
"soundness_avg": 2.2,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "3;3;3;3;2",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.495185"
} | {
"id": "HP7T95bcNm",
"metareview": "This paper proposes a watermarking method for latent diffusion models, which is denoted as WMAdapter with a contextual Adapter to fuse watermark information with the VAE features in the diffusion model. The paper proposes a hybrid finetuning strategy to further improve the visua... | {
"decision": "Reject"
} |
HAwZGLcye3 | 2405.17631v2 | BioDiscoveryAgent: An AI Agent for Designing Genetic Perturbation Experiments | {
"content": "## Abstract\n\nAbstract Agents based on large language models have shown great potential in accelerating scientific discovery by leveraging their rich background knowledge and reasoning capabilities. In this paper, we introduce BioDiscoveryAgent , an agent that designs new experiments, reasons about the... | [
{
"id": "HJN9q5HhD4",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces BioDiscoveryAgent,an AI-based tool for automating the design of genetic perturbation experiments. Leveraging large language models (LLMs) an... | {
"rating": "3;5;5;5;8",
"rating_avg": 5.2,
"confidence": "4;3;4;3;4",
"confidence_avg": 3.6,
"soundness": "3;3;3;3;4",
"soundness_avg": 3.2,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "2;3;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.495876"
} | {
"id": "56ArcLfwin",
"metareview": "The paper introduces BioDiscoveryAgent, an AI-based tool leveraging large language models (LLMs) to design genetic perturbation experiments, \"reason\" about their outcomes, and compare and contrast hypotheses explaining them. The agent identifies genes for perturbation to induc... | {
"decision": "Accept (Poster)"
} |
HB4lr0ykTi | 2411.00698v1 | Wasserstein Flow Matching: Generative modeling over families of distributions | {
"content": "## Abstract\n\nAbstract Generative modeling typically concerns the transport of a single source distribution to a single target distribution by learning (i.e., regressing onto) simple probability flows. However, in modern data-driven fields such as computer graphics and single-cell genomics, samples (sa... | [
{
"id": "UlLkW4xY0V",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a new type of generative model based on flow matching, i.e., Wasserstein flow matching. The proposed model allows for working with the *distri... | {
"rating": "3;5;8",
"rating_avg": 5.333333333333333,
"confidence": "3;4;3",
"confidence_avg": 3.3333333333333335,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;3;4",
"presentation_avg": 3.33333333333333... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.496713"
} | {
"id": "dLJRiU7PQV",
"metareview": "In the paper, the authors proposed Wasserstein flow matching, a type of generative model based on flow matching. While all the reviewers agree that the proposed method is novel, the paper still has some major weaknesses, including (1) the gap between theoretical justification an... | {
"decision": "Reject"
} |
HCJ7B6dhYK | 2410.19801v1 | Radon Implicit Field Transform (RIFT): Learning Scenes from Radar Signals | {
"content": "## Abstract\n\nAbstract Data acquisition in array signal processing (ASP) is costly, as high angular and range resolutions require large antenna apertures and wide frequency bandwidths. Data requirements grow multiplicatively with viewpoints and frequencies, increasing collection burdens. Implicit Neura... | [
{
"id": "u3Z0iYbE49",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 1,
"summary": "The paper presents the Radon Implicit Field Transform (RIFT), a novel method for reconstructing scenes from for synthetic aperture radar (SAR) imaging using an Im... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "4;3;2;2",
"contribution_avg": 2.75,
"presentation": "3;2;1;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.497448"
} | {
"id": "w2RfcVRL8R",
"metareview": "This paper proposes RIFT for SAR (Synthetic Aperture Radar), which is considered a special case of ASP (Array Signal Processing). Data acquisition for SAR, and ASP more generally, is notoriously challenging, and the authors posit that INR (Implicit Neural Representation) methods... | {
"decision": "Reject"
} |
HD6bWcj87Y | 2406.11011v2 | Data Shapley in One Training Run | {
"content": "## Abstract\n\nAbstract Data Shapley provides a principled framework for attributing data’s contribution within machine learning contexts. However, existing approaches require re-training models on different data subsets, which is computationally intensive, foreclosing their application to large-scale m... | [
{
"id": "kQ2DlDFSB3",
"initial_rating": 10,
"confidence": 3,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This paper introduces In-Run Data Shapley, a novel principled way of attributing the data contribution for deep learning models within one training run. A classi... | {
"rating": "5;6;8;10",
"rating_avg": 7.25,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "3;3;3;4",
"contribution_avg": 3.25,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.498320"
} | {
"id": "05S3iQCmd3",
"metareview": "The paper presents an approach aimed to evaluate the contributions of the data samples to the model. \nThis goal (attributing a relevance score to data samples w.r.t. the eventual model) is at the core of scientific, technical and legal concerns (privacy, copyright). \nThe metho... | {
"decision": "Accept (Oral)"
} |
HE6pJoNnFp | 2405.16178v1 | Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) augmented with retrieval exhibit robust performance and extensive versatility by incorporating external contexts.\nHowever, the input length grows linearly in the number of retrieved documents, causing a dramatic increase in latency.\nIn this paper, w... | [
{
"id": "wXI2Jv4gwn",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes a sparse RAG which consists of two stages – 1) document assessment, and 2) generation, to improve the efficiency of RAG. The document assessmen... | {
"rating": "3;5;6;6;8",
"rating_avg": 5.6,
"confidence": "4;3;3;4;4",
"confidence_avg": 3.6,
"soundness": "2;3;3;3;3",
"soundness_avg": 2.8,
"contribution": "2;3;3;2;3",
"contribution_avg": 2.6,
"presentation": "2;3;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.499280"
} | {
"id": "qutcUmihUS",
"metareview": "Summary of the paper: This paper introduces Sparse RAG designed to enhance the efficiency and effectiveness of traditional RAG. Sparse RAG addresses the challenges associated with increased memory usage of KV caches when processing lengthy documents, which can impede LLMs in man... | {
"decision": "Accept (Poster)"
} |
HGz012J6TQ | 2410.00327v1 | EnzymeFlow: Generating Reaction-specific Enzyme Catalytic Pockets through Flow Matching and Co-Evolutionary Dynamics | {
"content": "## Abstract\n\nAbstract Enzyme design is a critical area in biotechnology, with applications ranging from drug development to synthetic biology. Traditional methods for enzyme function prediction or protein binding pocket design often fall short in capturing the dynamic and complex nature of enzyme-subs... | [
{
"id": "Cfs3TX3BQg",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "EnzymeFlow introduces a new approach to conditionally generate enzyme pockets conditioned on the reaction of interest. They use a modified MultiFlow approach that... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "5;4;4;3",
"confidence_avg": 4,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "3;3;3;2",
"contribution_avg": 2.75,
"presentation": "3;4;2;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.500036"
} | {
"id": "5xElYBW2gN",
"metareview": "This paper seems to have several significant weaknesses. The most important are\n- The unclear role of the EC class as a conditioning factor. As mentioned by some reviewers, the actual source code suggests that MSA and EC class appear more like two multi-task learning tasks, wh... | {
"decision": "Reject"
} |
HJWdrvVyOi | 2201.10838v9 | Privacy-Preserving Logistic Regression Training with A Faster Gradient Variant | {
"content": "## Abstract\n\nAbstract Training logistic regression over encrypted data has been a compelling approach in addressing security concerns for several years. In this paper, we introduce an efficient gradient variant, called q u a d r a t i c 𝑞 𝑢 𝑎 𝑑 𝑟 𝑎 𝑡 𝑖 𝑐 quadratic italic_q ita... | [
{
"id": "QZHEKwCvYa",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The authors propose a privacy-preserving approach to training logistic regression based on homomorphic encryption. Their approach utilises second-order informatio... | {
"rating": "3;3;3;3;5",
"rating_avg": 3.4,
"confidence": "4;4;3;3;2",
"confidence_avg": 3.2,
"soundness": "2;3;2;2;3",
"soundness_avg": 2.4,
"contribution": "1;2;2;2;3",
"contribution_avg": 2,
"presentation": "2;2;3;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.500902"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
HJp1g4w1Or | 2410.19986v1 | Resolving Domain Shift For Representations Of Speech In Non-Invasive Brain Recordings | {
"content": "## Abstract\n\nAbstract Machine learning techniques have enabled researchers to leverage neuroimaging data to decode speech from brain activity, with some amazing recent successes achieved by applications built using invasive devices. However, research requiring surgical implants has a number of practic... | [
{
"id": "gjlEzZyK7r",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "Authors present the first deep MEG dataset alignment/adaptation method, albeit a common goal outside of the deep learning methods and MEG modality. For instance, ... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "1;2;3;3",
"soundness_avg": 2.25,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "2;3;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.501586"
} | {
"id": "viiinB1Jk5",
"metareview": "This submission contributes adverserial MEG adaptation methods. The work generated interest and discussion with the reviewers. However, it fail to convince that it met the high bar for acceptance at ICLR. In particular, the reviewers felt that there was insufficient positioning ... | {
"decision": "Reject"
} |
HMVDiaWMwM | 2406.01561v3 | Guided Score Identity Distillation for Data-Free One-Step Text-to-Image Generation | {
"content": "## Abstract\n\nAbstract Diffusion-based text-to-image generation models trained on extensive text-image pairs have shown the capacity to generate photorealistic images consistent with textual descriptions. However, a significant limitation of these models is their slow sample generation, which requires ... | [
{
"id": "XRehdoDeLj",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper focuses on a novel approach to accelerating text-to-image diffusion models through Score Identity Distillation with Long-Short Guidance (SiD-LSG), spec... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "2;2;3;4",
"confidence_avg": 2.75,
"soundness": "4;3;3;4",
"soundness_avg": 3.5,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"presentation": "3;2;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.502454"
} | {
"id": "Hvn1bIS4EZ",
"metareview": "This paper focuses on accelerating diffusion-based generative models. It introduces a novel method, Guided Score Identity Distillation with Long-Short Guidance (SiD-LSG), designed for data-free one-step text-to-image generation. The approach builds upon the existing Score Identi... | {
"decision": "Accept (Poster)"
} |
HMrcv7Q4Ub | 2410.23317v1 | VL-Cache: Sparsity and Modality-Aware KV Cache Compression for Vision-Language Model Inference Acceleration | {
"content": "## Abstract\n\nAbstract Vision-Language Models (VLMs) have demonstrated impressive performance across a versatile set of tasks. A key challenge in accelerating VLMs is storing and accessing the large Key-Value (KV) cache that encodes long visual contexts, such as images or videos. While existing KV cach... | [
{
"id": "Yeqyd4CEjq",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper addresses the challenge of efficiently storing and accessing large Key-Value (KV) caches in Vision-Language Models (VLMs), which are crucial for handli... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;5;4;4",
"confidence_avg": 4.25,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.503202"
} | {
"id": "9oNOYiabEp",
"metareview": "The paper introduces VL-Cache, a novel KV cache compression method tailored for accelerating vision-language model inference, which significantly reduces memory usage and latency while maintaining high accuracy.\n\nAfter rebuttal and discussions, this paper receives all positive... | {
"decision": "Accept (Poster)"
} |
HN8V0flwJF | 2402.08268v3 | World Model on Million-Length Video And Language With Blockwise RingAttention | {
"content": "## Abstract\n\nAbstract Current language models fall short in understanding aspects of the world not easily described in words, and struggle with complex, long-form tasks. Video sequences offer valuable temporal information absent in language and static images, making them attractive for joint modeling ... | [
{
"id": "DcRdt5IzmK",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "This paper proposes a novel model architecture and training process that achieves significant advancements in long-context modeling for developing large language ... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "5;5;5;4",
"confidence_avg": 4.75,
"soundness": "2;4;4;3",
"soundness_avg": 3.25,
"contribution": "2;4;2;4",
"contribution_avg": 3,
"presentation": "2;3;4;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.503852"
} | {
"id": "lJRsQap9x4",
"metareview": "### Summary:\nThis paper introduces LWM (Large World Model), a novel approach for processing million-length sequences in both language and video modalities. The key contribution is demonstrating scalable training of models that can handle extremely long context windows (up to 1M... | {
"decision": "Accept (Poster)"
} |
HSGCCUwH7r | 2410.11163v1 | Model Swarms: Collaborative Search to Adapt LLM Experts via Swarm Intelligence | {
"content": "## Abstract\n\nAbstract We propose Model Swarms , a collaborative search algorithm to adapt LLMs via swarm intelligence , the collective behavior guiding individual systems. Specifically, Model Swarms starts with a pool of LLM experts and a utility function. Guided by the best-found checkpoints across m... | [
{
"id": "l1jFTWWqD6",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes a collaborative search algorithm that uses swarm intelligence to adapt LLMs. The method proposed starts with different LLM experts and a utili... | {
"rating": "3;5;8;8",
"rating_avg": 6,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;3;4;4",
"soundness_avg": 3.5,
"contribution": "2;2;4;3",
"contribution_avg": 2.75,
"presentation": "3;3;4;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.504586"
} | {
"id": "SBW38HO1RC",
"metareview": "This paper seems to explore an interesting concept but the main concern here is that its technical narrative is not clear. There is very little explanation of the main algorithm.\n\nFor example, in lines 122-123, there is a non-trivial form for the velocity update that is hard t... | {
"decision": "Reject"
} |
HSi4VetQLj | 2406.04306v1 | Improving Uncertainty Estimation through Semantically Diverse Language Generation | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) can suffer from hallucinations when generating text. These hallucinations impede various applications in society and industry by making LLMs untrustworthy. Current LLMs generate text in an autoregressive fashion by predicting and appending text tokens... | [
{
"id": "LlEoG2BPXL",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces Semantically Diverse Language Generation (SDLG) to enhance uncertainty measurement by generating semantically diverse output sequences. SDLG... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "2;4;4;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;2;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.505478"
} | {
"id": "WJqRyYKHf2",
"metareview": "The paper presents a method called SLDG to enhance uncertainty measurement in language generation models. The novelty lies in using semantic divergence of the generated sequences to measure uncertainty. The method results in better uncertainty estimation than SOTA. Other than... | {
"decision": "Accept (Poster)"
} |
HVJBFYJrN2 | 2407.10935v1 | STARS: Self-supervised Tuning for 3D Action Recognition in Skeleton Sequences | {
"content": "## Abstract\n\nAbstract Self-supervised pretraining methods with masked prediction demonstrate remarkable within-dataset performance in skeleton-based action recognition. However, we show that, unlike contrastive learning approaches, they do not produce well-separated clusters. Additionally, these metho... | [
{
"id": "ZreKAhPTAn",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "- The paper tackles the problem of self-supervised learning for skeleton-representation learning. \n- While previous state-of-the-art are MAE-based and achieve go... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;5;4;5",
"confidence_avg": 4.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.506180"
} | {
"id": "qgrJ7roXNA",
"metareview": "This paper introduces a human behaviour analysis method, STARS, aiming to enhance the output representation of MAE encoders. However, this work lacks sufficient novelty, and it seems to be a stack of existing techniques, like MAE and contrastive learning methods, to build the pr... | {
"decision": "Reject"
} |
HYaUHZAoPc | 2410.07525v2 | Offline Inverse Constrained Reinforcement Learning for Safe-Critical Decision Making in Healthcare | {
"content": "## Abstract\n\nAbstract Reinforcement Learning (RL) applied in healthcare can lead to unsafe medical decisions and treatment, such as excessive dosages or abrupt changes, often due to agents overlooking common-sense constraints. Consequently, Constrained Reinforcement Learning (CRL) is a natural choice ... | [
{
"id": "Dt0pwkCDL8",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposed a constraint transformer method to tackle the problem that only historical data is provided in offline inverse reinforcement learning. The prop... | {
"rating": "3;3;5;5;6",
"rating_avg": 4.4,
"confidence": "5;3;4;5;2",
"confidence_avg": 3.8,
"soundness": "2;3;3;3;3",
"soundness_avg": 2.8,
"contribution": "2;3;2;3;2",
"contribution_avg": 2.4,
"presentation": "2;3;2;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.506814"
} | {
"id": "446T9g80ZH",
"metareview": "This paper proposes a transformer based model to overcome some challenges with safety in healthcare RL by using the decision-transformer framework combined with a reward design (IRL) setup where markovian assumptions may be violated. Empirical evaluation compares results in sep... | {
"decision": "Reject"
} |
HZVIQE1MsJ | 2410.03742v2 | Learning Generative Judge from Preference Data | {
"content": "## Abstract\n\nAbstract Learning from preference feedback is a common practice for aligning large language models (LLMs) with human value.\nConventionally, preference data is learned and encoded into a scalar reward model that connects a value head with an LLM to produce a scalar score as preference or ... | [
{
"id": "qrrUVcUsqM",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes a generative approach to build reward models for preference learning. They generate judgements with explanations for both positive and negative... | {
"rating": "6;6;6;6",
"rating_avg": 6,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;4;3;3",
"contribution_avg": 3,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.507584"
} | {
"id": "fQOYwAclvv",
"metareview": "This paper introduces Con-J, a method for aligning LLMs with human values that uses a generative LLM to produce contrastive judgments with natural language rationales, addressing the limitations of traditional scalar reward models. Training this \"Judge\" with Direct Preference ... | {
"decision": "Accept (Poster)"
} |
HZxJfzs3w6 | 2410.19814v1 | Stochastic Flow Matching for Resolving Small-Scale Physics | {
"content": "## Abstract\n\nAbstract Conditioning diffusion and flow models have proven effective for super-resolving small-scale details in natural images. However, in physical sciences such as weather, super-resolving small-scale details poses significant challenges due to: ( i ) 𝑖 (i) ( italic_i ) misalignment b... | [
{
"id": "FsRM2duOAL",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces the Stochastic Flow Matching (SFM) framework, a novel approach to super-resolve small-scale physics in meteorological data, particularly in d... | {
"rating": "5;5;5;5;6",
"rating_avg": 5.2,
"confidence": "4;3;3;3;2",
"confidence_avg": 3,
"soundness": "3;3;2;2;3",
"soundness_avg": 2.6,
"contribution": "3;2;2;2;3",
"contribution_avg": 2.4,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.508287"
} | {
"id": "0w5xYg5LX5",
"metareview": "The paper introduces an Adaptive Flow Matching (AFM) framework for super-resolving small-scale physics, but reviewers raised concerns about its novelty and positioning relative to existing methods like CorrDiff and diffusion-based approaches. A key weakness highlighted is the la... | {
"decision": "Reject"
} |
HZz81oCNlp | 2405.17074v1 | Towards Ultra-High-Definition Image Deraining: A Benchmark and An Efficient Method | {
"content": "## Abstract\n\nAbstract Despite significant progress has been made in image deraining, existing approaches are mostly carried out on low-resolution images. The effectiveness of these methods on high-resolution images is still unknown, especially for ultra-high-definition (UHD) images, given the continuo... | [
{
"id": "ncwPlsO0Lt",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper addresses the task of deraining UHD images, which is not yet well-explored despite advances in low-resolution image deraining research. The authors int... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "5;5;3;5",
"confidence_avg": 4.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.509045"
} | {
"id": "i0bJI5xSNg",
"metareview": "This paper presents the first dataset for UHD-resolution image draining and a computationally efficient deraining model for such high-resolution images. The dataset outstands the existing lower-resolution datasets as the rain streak shape patterns differ in higher resolutions. T... | {
"decision": "Reject"
} |
Ha6RTeWMd0 | 2408.00714v2 | SAM 2: Segment Anything in Images and Videos | {
"content": "## Abstract\n\nAbstract We present Segment Anything Model 2 (SAM 2), a foundation model\ntowards solving promptable visual segmentation in images and videos . We build a data engine, which improves model and data via user interaction, to collect the largest video segmentation dataset to date. Our model ... | [
{
"id": "YTu5kFXmZ4",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "In this paper, the authors build a data engine to generate a large-scale video segmentation dataset. Using the datasets, they train a strong yet efficient model."... | {
"rating": "8;8;8;8",
"rating_avg": 8,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "4;3;4;4",
"soundness_avg": 3.75,
"contribution": "4;3;4;3",
"contribution_avg": 3.5,
"presentation": "4;3;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.510007"
} | {
"id": "fxbBSPSQCH",
"metareview": "This work presents a visual foundation model, SAM 2, achieving new state-of-the-art results on video/image interactive segmentation and video object segmentation. \n\nSeveral issues in the earlier version are about missing compared methods, close related works on video semantic-... | {
"decision": "Accept (Oral)"
} |
HaX48yksVL | 2406.17216v1 | Machine Unlearning Fails to Remove Data Poisoning Attacks | {
"content": "## Abstract\n\nAbstract We revisit the efficacy of several practical methods for approximate machine unlearning developed for large-scale deep learning. In addition to complying with data deletion requests, one often-cited potential application for unlearning methods is to remove the effects of training... | [
{
"id": "8tUuFlnnOw",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper evaluates the efficacy of several machine unlearning methods across several tasks, including one introduced by the authors: Gaussian poisoning.",
"s... | {
"rating": "5;5;5;5;6",
"rating_avg": 5.2,
"confidence": "3;4;3;4;4",
"confidence_avg": 3.6,
"soundness": "2;2;3;3;3",
"soundness_avg": 2.6,
"contribution": "3;3;3;3;2",
"contribution_avg": 2.8,
"presentation": "2;2;2;3;3",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.511203"
} | {
"id": "td4QhvSYk9",
"metareview": "After thorough consideration of the authors' rebuttal and the reviewers' feedback, a positive consensus has been reached. The authors have effectively addressed several critical concerns raised in the initial reviews, demonstrating a commitment to improving the quality and clari... | {
"decision": "Accept (Poster)"
} |
HaXlWs1LX8 | 2410.16579v1 | Conflict-Aware Adversarial Training | {
"content": "## Abstract\n\nAbstract Adversarial training is the most effective method to obtain adversarial robustness for deep neural networks by directly involving adversarial samples in the training procedure. To obtain an accurate and robust model, the weighted-average method is applied to optimize standard los... | [
{
"id": "1YdMepndq6",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper explores the issue of gradient conflict in adversarial training, highlighting how traditional methods struggle to balance standard performance and adver... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.512105"
} | {
"id": "HJYaFYUvoT",
"metareview": "The paper tackles an important problem in adversarial training by addressing gradient conflicts through a conflict-aware trade-off factor. However, the proposed method lacks sufficient novelty, as gradient alignment has been explored in existing works, and CA-AT appears more inc... | {
"decision": "Reject"
} |
HafxTJjo6a | 2410.24151v1 | Scaling Concept With Text-Guided Diffusion Models | {
"content": "## Abstract\n\nAbstract Text-guided diffusion models have revolutionized generative tasks by producing high-fidelity content from text descriptions. They have also enabled an editing paradigm where concepts can be replaced through text conditioning (e.g., a dog → → \\rightarrow → a tiger ). In this work... | [
{
"id": "7I9TUrimOT",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents ScalingConcept which aims to enhance or suppress existing concepts in real input data using text-guided diffusion models. The proposed method l... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "3;4;5;3",
"confidence_avg": 3.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.512761"
} | {
"id": "LAhaVmIwnI",
"metareview": "- Scientific Claims and Findings:\n - This paper presents a new concept scaling method for text-to-image generation. The approach allows for the enhancement or suppression of concepts in input images by adjusting the noise difference between concept reconstruction and concept... | {
"decision": "Reject"
} |
HbZrxBXzks | 2408.10120v1 | Geometry Informed Tokenization of Molecules for Language Model Generation | {
"content": "## Abstract\n\nAbstract We consider molecule generation in 3D space using language models (LMs), which requires discrete tokenization of 3D molecular geometries. Although tokenization of molecular graphs exists, that for 3D geometries is largely unexplored. Here, we attempt to bridge this gap by proposi... | [
{
"id": "ShNE3f7La2",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes Geo2Seq tokenization scheme that is meant to enable generation of 3D molecules using autoregressive LLMs.\nIt canonicalizes the graph and crea... | {
"rating": "3;5;5;6;6;6;6;10",
"rating_avg": 5.875,
"confidence": "3;5;4;4;3;3;2;3",
"confidence_avg": 3.375,
"soundness": "2;3;2;3;3;4;4;4",
"soundness_avg": 3.125,
"contribution": "2;2;2;2;3;3;2;3",
"contribution_avg": 2.375,
"presentation": "2;3;2;3;4;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.514103"
} | {
"id": "S3bvYu2or8",
"metareview": "**Summary**\nThis work proposes a geometry-informed tokenization method Geo2Seq to enable 3D molecule generation with language models (LMs). Specifically, Geo2Seq employs canonical labeling and invariant spherical representations to transform, reversibly, 3D structures into SE(3... | {
"decision": "Reject"
} |
HbbnlrmsAH | 2410.10469v1 | Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts | {
"content": "## Abstract\n\nAbstract Time series foundation models have demonstrated impressive performance as zero-shot forecasters, i.e., they can tackle a wide variety of downstream forecasting tasks without explicit task-specific training. However, achieving effectively unified training on time series remains an... | [
{
"id": "CcRI0EUXpX",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces Moirai-MoE, a method for unified time series forecasting, which addresses challenges of the human-imposed frequency-level model specializati... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;5;3;4",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.515452"
} | {
"id": "nMY6Fwbls5",
"metareview": "(a) Summary of Scientific Claims and Findings\nThis paper proposes Moirai-MoE, a sparse mixture of experts (MoE) framework designed to address the challenges of unified training for time series foundation models. The key contributions include:\nA token-level specialization mecha... | {
"decision": "Reject"
} |
Hbpzrh7JbN | 2405.17066v1 | Saturn: Sample-efficient Generative Molecular Design using Memory Manipulation | {
"content": "## Abstract\n\nAbstract Generative molecular design for drug discovery has very recently achieved a wave of experimental validation, with language-based backbones being the most common architectures employed. The most important factor for downstream success is whether an in silico oracle is well correla... | [
{
"id": "RbGLsSDOzu",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors present Saturn, an RL-approach using a Mamba backbone for generative molecular design. Design choices surrounding implementation are thoroughly explor... | {
"rating": "3;3;5;6;6",
"rating_avg": 4.6,
"confidence": "3;4;3;2;3",
"confidence_avg": 3,
"soundness": "2;4;3;3;3",
"soundness_avg": 3,
"contribution": "2;1;3;2;3",
"contribution_avg": 2.2,
"presentation": "2;4;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.516296"
} | {
"id": "8pQkAp35ua",
"metareview": "This submission introduces Saturn for generative molecular design based on the language model backbone Mamba. Saturn integrates augmented memory in RL for more efficient generation with strong empirical performance in drug discovery. \n\nThe reviewers agreed that the performed e... | {
"decision": "Reject"
} |
Hcb2cgPbMg | 2406.06811v2 | Learning Continually by Spectral Regularization | {
"content": "## Abstract\n\nAbstract † † footnotetext: Correspondence to: Alex Lewandowski <lewandowski@ualberta.ca>. Loss of plasticity is a phenomenon where neural networks can become more difficult to train over the course of learning.\nContinual learning algorithms seek to mitigate this effect by sustaining good... | [
{
"id": "VCBpYoCW3w",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The authors derive a new spectral regularizer for continual learning inspired by the observation that the singular values of the neural network parameters at init... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.517326"
} | {
"id": "2jJ2w7xH7z",
"metareview": "The paper proposes spectral regularization scheme for continual learning tasks. Instead of simple L2 regularization, the proposed scheme aims to regularize the largest singular value of the weight matrices close to 1. While spectral regularization has been widely used for improv... | {
"decision": "Accept (Poster)"
} |
HgSIfXTpBE | 2406.05227v2 | Mixed-curvature decision trees and random forests | {
"content": "## Abstract\n\nAbstract We extend decision tree and random forest algorithms to mixed-curvature product spaces.\nSuch spaces, defined as Cartesian products of Euclidean, hyperspherical, and hyperbolic manifolds, can often embed points from pairwise distances with much lower distortion than in single man... | [
{
"id": "8FWPPXcVfi",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "Summary\nThis paper proposes to extend decision trees (DT) and random forests (RF) to mixed curvature spaces -- meaning that the data (here assumed to be in ambie... | {
"rating": "3;5;5;6;8",
"rating_avg": 5.4,
"confidence": "4;3;4;4;3",
"confidence_avg": 3.6,
"soundness": "2;3;2;3;4",
"soundness_avg": 2.8,
"contribution": "3;2;2;3;3",
"contribution_avg": 2.6,
"presentation": "2;4;3;3;4",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.518382"
} | {
"id": "AlqihsfSrF",
"metareview": "The authors propose to extend decision trees (DTs) and random forest (RF) into the product manifolds of hyperbolic, hyperspherical, or Euclidean components with a general angular split. The authors raised concerns about its novelty given results on product manifolds, and DTs/RF ... | {
"decision": "Reject"
} |
Hh6XKefS28 | 2407.02779v1 | Croppable Knowledge Graph Embedding | {
"content": "## Abstract\n\nAbstract Knowledge Graph Embedding (KGE) is a common method for Knowledge Graphs (KGs) to serve various artificial intelligence tasks. The suitable dimensions of the embeddings depend on the storage and computing conditions of the specific application scenarios. Once a new dimension is re... | [
{
"id": "oOjfA7TTHl",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "Knowledge Graph Embeddings (KGEs) project entities and relationships into a continuous vector space and are widely applied in tasks like link prediction. Typicall... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.519009"
} | {
"id": "aRGhc8jhba",
"metareview": "This manuscript proposes a Knowledge Graph Embedding (KGE) training framework named MED to address the challenge of efficiently adapting KGE models to various dimensional requirements without retraining from scratch. MED consists of three main modules: the Mutual learning mechan... | {
"decision": "Reject"
} |
HhefvT4ktU | 2402.01002v2 | AI-generated faces influence gender stereotypes and racial homogenization | {
"content": "## Abstract\n\nAbstract Text-to-image generative AI models such as Stable Diffusion are used daily by millions worldwide. However, the extent to which these models exhibit racial and gender stereotypes is not yet fully understood. Here, we document significant biases in Stable Diffusion across six races... | [
{
"id": "nVsYCAPZxC",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes debiasing AI models ( Stable Diffusion) by developing their classifiers with more inclusive datasets and then uses those inclusive images to as... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;4;5;3",
"confidence_avg": 4,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "2;2;2;1",
"contribution_avg": 1.75,
"presentation": "2;2;1;1",
"presentation_avg": 1.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.519649"
} | {
"id": "QVu8nzNvPI",
"metareview": "The paper proposes debiasing Stable Diffusion by developing classifiers with more inclusive datasets. To this end, the authors have curated a SoTA classifier for race, gender, profession from faces and show that the majority of faces generated by SDXL are white males and exhibit... | {
"decision": "Reject"
} |
Hhx3swAQAZ | 2406.14130v1 | ExVideo: Extending Video Diffusion Models via Parameter-Efficient Post-Tuning | {
"content": "## Abstract\n\nAbstract. Recently, advancements in video synthesis have attracted significant attention. Video synthesis models such as AnimateDiff and Stable Video Diffusion have demonstrated the practical applicability of diffusion models in creating dynamic visual content. The emergence of SORA has f... | [
{
"id": "TdsHY2yOjs",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a post-tuning method to enhance current video generation models, allowing them to produce longer videos with lower training costs. It levera... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;1;2;3",
"soundness_avg": 2,
"contribution": "1;1;2;3",
"contribution_avg": 1.75,
"presentation": "2;2;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.520265"
} | {
"id": "QugU8DOegb",
"metareview": "Reviewers unanimously recommend rejection of the paper, citing lack of novelty, clarity in presentation, lack of comparisons, among other things. \n\nAuthors did not participate in discussion phase, and as such, the initial scores persist. \n\nThere are a handful of good bits of... | {
"decision": "Reject"
} |
HkB4bW5eJj | 2410.02637v1 | Plots unlock time-series understanding in multimodal models | {
"content": "## Abstract\n\nAbstract While multimodal foundation models can now natively work with data beyond text, they remain underutilized in analyzing the considerable amounts of multi-dimensional time-series data in fields like healthcare, finance, and social sciences, representing a missed opportunity for ric... | [
{
"id": "i4NHTkb2e6",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 4,
"presentation": 2,
"summary": "This paper demonstrates that for some synthetic and real-world time series understanding problems (for example, detecting the functional form of a time series, an... | {
"rating": "1;3;5;8",
"rating_avg": 4.25,
"confidence": "2;5;3;4",
"confidence_avg": 3.5,
"soundness": "1;2;2;3",
"soundness_avg": 2,
"contribution": "1;4;2;3",
"contribution_avg": 2.5,
"presentation": "1;2;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.520859"
} | {
"id": "tUp7gUZnw5",
"metareview": "This paper has been evaluated by 4 knowledgeable reviewers and their initial opinions varied: 1 strong rejection, 1 marginal rejection and 1 marginal acceptance. The authors provided extensive rebuttals but that did not help the scores. Even though the key concept presented in t... | {
"decision": "Reject"
} |
Hlm0cga0sv | 2411.07199v1 | OmniEdit: Building Image Editing Generalist Models Through Specialist Supervision | {
"content": "## Abstract\n\nAbstract Instruction-guided image editing methods have demonstrated significant potential by training diffusion models on automatically synthesized or manually annotated image editing pairs. However, these methods remain far from practical, real-life applications. We identify three primar... | [
{
"id": "j4z43I9do4",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "OMNI-EDIT is an image editing model designed to address the skill imbalance issues present in existing instruction-based image editing methods. It learns by combi... | {
"rating": "5;5;5;6;6",
"rating_avg": 5.4,
"confidence": "3;5;4;3;4",
"confidence_avg": 3.8,
"soundness": "3;3;3;2;3",
"soundness_avg": 2.8,
"contribution": "3;2;2;2;3",
"contribution_avg": 2.4,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.521562"
} | {
"id": "VIikEtQAiK",
"metareview": "This paper presents an instruction-based image editing method for seven different tasks, trained on the basis of pre-trained task-specific models, as well as LLM scoring. A novel architecture is presented and trained using images with different image resolutions. The authors rel... | {
"decision": "Accept (Poster)"
} |
HmwneoGoy9 | 2410.13276v2 | SeerAttention: Learning Intrinsic Sparse Attention in Your LLMs | {
"content": "## Abstract\n\nAbstract Attention is the cornerstone of modern Large Language Models (LLMs). Yet its quadratic complexity limits the efficiency and scalability of LLMs, especially for those with a long-context window. A promising approach addressing this limitation is to leverage the sparsity in attenti... | [
{
"id": "oaTZqV63Iw",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work introduces SeerAttention, which uses pooled embedding to compute attention block mask and uses the mask to do block sparse attention for reduced latency... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "4;3;2;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.522241"
} | {
"id": "bwCQRUgxLb",
"metareview": "**Summary:** The paper introduces SeerAttention, which exploits block-level sparsity in attention maps by learning a gating function. After employing a gating network with block-wise pooling and a customized FlashAttention kernel, the approach can achieve up to 5.67x speedups (f... | {
"decision": "Reject"
} |
HnhNRrLPwm | 2410.10139v1 | MMIE: Massive Multimodal Interleaved Comprehension Benchmark for Large Vision-Language Models | {
"content": "## Abstract\n\nAbstract Interleaved multimodal comprehension and generation, enabling models to produce and interpret both images and text in arbitrary sequences, have become a pivotal area in multimodal learning. Despite significant advancements, the evaluation of this capability remains insufficient. ... | [
{
"id": "lXzoRAuR26",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper presents MMIE: a benchmark for evaluating interleaved multimodal comprehension and generation abilities of Multimodal LLMs. The evaluation dataset is al... | {
"rating": "6;8;8",
"rating_avg": 7.333333333333333,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "3;3;4",
"soundness_avg": 3.3333333333333335,
"contribution": "3;2;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;4",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.522807"
} | {
"id": "Q6MtbPVVUC",
"metareview": "This paper introduces MMIE (Massive Multimodal Interleaved Evaluation), a benchmark for evaluating interleaved multimodal comprehension and generation abilities of Multimodal LLMs. MMIE comprises 20K multimodal queries across 12 fields, supporting interleaved text and multi-imag... | {
"decision": "Accept (Oral)"
} |
HoyKFRhwMS | 2408.08172v2 | Towards flexible perception with visual memory | {
"content": "## Abstract\n\nAbstract Training a neural network is a monolithic endeavor, akin to carving knowledge into stone: once the process is completed, editing the knowledge in a network is nearly impossible, since all information is distributed across the network’s weights. We here explore a simple, compellin... | [
{
"id": "d7MZQtgtYP",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "The authors address the memory-based image classification task by utilizing large, unsupervised pretrained visual models to extract image descriptors. They perfor... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "4;3;5;4",
"confidence_avg": 4,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "1;3;3;3",
"contribution_avg": 2.5,
"presentation": "3;4;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.523493"
} | {
"id": "PIwJLfwjMO",
"metareview": "This work introduces a flexible and interpretable visual memory system that combines deep neural network embeddings with database-like capabilities, enabling scalable data addition, unlearning, and intervention for controlled decision-making.\n\nThe paper is generally well-writt... | {
"decision": "Reject"
} |
Hpu3KIX8Am | 2406.02549v1 | Dreamguider: Improved Training free Diffusion-based Conditional Generation | {
"content": "## Abstract\n\nAbstract Diffusion models have emerged as a formidable tool for training-free conditional generation.\nHowever, a key hurdle in inference-time guidance techniques is the need for compute-heavy backpropagation through the diffusion network for estimating the guidance direction. Moreover, t... | [
{
"id": "R8sezLGITO",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 1,
"presentation": 1,
"summary": "The authors propose a generic lightweight guidance solution, named Dreamguider, which enables inference-time guidance without the need for backpropagation through... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;5;2;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;2",
"soundness_avg": 2.5,
"contribution": "2;1;3;3",
"contribution_avg": 2.25,
"presentation": "2;1;3;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.524110"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
HqLHY4TzGj | 2311.18512v1 | Union-over-Intersections: Object Detection beyond Winner-Takes-All | {
"content": "## Abstract\n\nAbstract This paper revisits the pipeline for detecting objects in images with proposals. For any object detector, the obtained box proposals or queries need to be classified and regressed towards ground truth boxes. The common solution for the final predictions is to directly maximize th... | [
{
"id": "QWrsfeqjrN",
"initial_rating": 8,
"confidence": 5,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This paper revisits the problem of predicting box locations in object detection architectures. Traditional approaches typically regress box proposals to maximize ... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "5;4;4;5",
"confidence_avg": 4.5,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;2;4",
"contribution_avg": 2.5,
"presentation": "2;4;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.524660"
} | {
"id": "awY1ynfiHM",
"metareview": "(a) The authors propose the Union-over-Intersections (UoI) method, which improves object detection by focusing regression on intersection areas for better localization and combining proposal information through union in post-processing.\n\n(b) The strengths of the paper include ... | {
"decision": "Accept (Spotlight)"
} |
HqlX3lPtbh | 2312.04403v1 | OT-Attack: Enhancing Adversarial Transferability of Vision-Language Models via Optimal Transport Optimization | {
"content": "## Abstract\n\nAbstract Vision-language pre-training (VLP) models demonstrate impressive abilities in processing both images and text. However, they are vulnerable to multi-modal adversarial examples (AEs). Investigating the generation of high-transferability adversarial examples is crucial for uncoveri... | [
{
"id": "vWRUUY06lZ",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper focuses on the task of conducting adversarial attacks on vision-language models. The focus of the attack is transferability. The paper introduces the o... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;5;3;3",
"confidence_avg": 3.75,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.525338"
} | {
"id": "IDOUq3ks9Y",
"metareview": "This paper focuses on the task of conducting adversarial attacks on vision-language models. The focus of the attack is transferability. The paper introduces the optimal transport perspective for finding more transferable attacks by considering the optimal alignment between augme... | {
"decision": "Reject"
} |
Hr3TBaZl4S | 2410.15698v1 | Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces | {
"content": "## Abstract\n\nAbstract Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories.\nHowever, most research only focuses on limited continual task settings where the tasks have the same obs... | [
{
"id": "S3kLVEqUZS",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a method for the problem of continual offline reinforcement learning. The proposed method combines two modules: the Quantized Spaces Alignme... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "3;1;3;2",
"soundness_avg": 2.25,
"contribution": "1;2;3;2",
"contribution_avg": 2,
"presentation": "1;1;2;3",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.526055"
} | {
"id": "bHuXvCWhnW",
"metareview": "**summary**\n\nThis paper introduces Vector-Quantized Continual Diffuser (VQ-CD), a novel method for continual offline RL, capable of handling tasks with differing observation and action spaces. VQ-CD integrates two core modules: Quantized Spaces Alignment (QSA), which uses vect... | {
"decision": "Reject"
} |
HrdVqFSn1e | 2410.14237v1 | Unified Convergence Analysis for Score-Based Diffusion Models with Deterministic Samplers | {
"content": "## Abstract\n\nAbstract Score-based diffusion models have emerged as powerful techniques for generating samples from high-dimensional data distributions. These models involve a two-phase process: first, injecting noise to transform the data distribution into a known prior distribution, and second, sampl... | [
{
"id": "6xlACtdWpE",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "The paper addresses a previously unexplored question: bounding the total variation between the true distribution and the distribution generated by an ODE sampler ... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "2;2;3;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.527449"
} | {
"id": "EpVUOsfo5x",
"metareview": "This paper introduces a unified framework that analyzed several existing first-order ODE-based samplers, leading to an iteration complexity of $O(d^2/\\epsilon)$. The unified analysis may benefit future development of convergence rates in a modular manner, and hence received pos... | {
"decision": "Accept (Poster)"
} |
Hs1UTIOwKr | 2410.01288v2 | Mitigating Copy Bias in In-Context Learning through Neuron Pruning | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have demonstrated impressive few-shot in-context learning (ICL) abilities. Still, we show that they are sometimes prone to a ‘copying bias’, where they copy answers from provided examples instead of learning the underlying patterns. In this work, we p... | [
{
"id": "dAUfTlDmAp",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents a novel approach to addressing a key issue in large language models (LLMs) during In-Context Learning (ICL)—the tendency to copy answers from p... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.529472"
} | {
"id": "iYqBg51lXS",
"metareview": "The paper proposes pruning neurons responsible for copying bias in few-shot in-context learning, identified via Integrated Gradients. While the authors report some improvements on synthetic tasks, the overall scope and novelty are limited. Key details are missing, and the paper ... | {
"decision": "Reject"
} |
HsB1sQvXML | 2409.03291v2 | LLM Detectors Still Fall Short of Real World: Case of LLM-Generated Short News-Like Posts | {
"content": "## Abstract\n\nAbstract With the emergence of widely available powerful LLMs, disinformation generated by large Language Models (LLMs) has become a major concern. Historically, LLM detectors have been touted as a solution, but their effectiveness in the real world is still to be proven. In this paper, w... | [
{
"id": "8SPpOvjVvG",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper examines the effectiveness of existing large language model (LLM) detectors in identifying short news-like posts generated by LLMs in real-world scenar... | {
"rating": "3;3;3;5;5",
"rating_avg": 3.8,
"confidence": "3;3;4;4;4",
"confidence_avg": 3.6,
"soundness": "2;2;1;3;3",
"soundness_avg": 2.2,
"contribution": "2;2;1;2;2",
"contribution_avg": 1.8,
"presentation": "1;2;2;3;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.530317"
} | {
"id": "FUjBsVhdYb",
"metareview": "The reviewers unanimously voted to reject this paper, and I tend to agree with the reviewers here. The reviewers point out that the contributions here are modest, the experiments are narrow, the presentation is lacking, among other criticisms. The reviewers do believe this pap... | {
"decision": "Reject"
} |
HtS2NUcqtD | 2405.16411v2 | Tensor Attention Training: Provably Efficient Learning of Higher-order Transformers | {
"content": "## Abstract\n\nTensor Attention, a multi-view attention that is able to capture high-order correlations among multiple modalities, can overcome the representational limitations of classical matrix attention. However, the O ( n 3 ) 𝑂 superscript 𝑛 3 O(n^{3}) italic_O ( italic_n start_POSTSUPERSCRIPT ... | [
{
"id": "GVhDSZxcEo",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "This paper derives an efficient algorithm to approximate the gradients of high-order transformers. Such models extend the classical self attention mechanism, wher... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "3;4;2;4",
"confidence_avg": 3.25,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;1;3;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.531563"
} | {
"id": "aB5SfMltBN",
"metareview": "This paper gives an algorithm for approximating the gradient for tensor attention in n^{1+o(1)} time under certain bounds on the weights, and showed a lowerbound when then weights are not bounded. The reviewers have a large disagreement on the evaluation of this paper. Most revi... | {
"decision": "Reject"
} |
HtbqsbNw9c | 2405.12069v2 | Gaussian Head & Shoulders: High Fidelity Neural Upper Body Avatars with Anchor Gaussian Guided Texture Warping | {
"content": "## Abstract\n\nAbstract. The ability to reconstruct realistic and controllable upper body avatars from casual monocular videos is critical for various applications in communication and entertainment. By equipping the most recent 3D Gaussian Splatting representation with head 3D morphable models (3DMM), ... | [
{
"id": "nUUmPZ7pHk",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a method to reconstruct upper body avatars from videos based on 3D Gaussian Splatting. The method models head and shoulders separately, i.e., ... | {
"rating": "3;5;5;6;6",
"rating_avg": 5,
"confidence": "4;3;3;3;4",
"confidence_avg": 3.4,
"soundness": "2;2;3;4;3",
"soundness_avg": 2.8,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "3;3;1;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.532823"
} | {
"id": "RdUU1PlydN",
"metareview": "This paper proposes a method for constructing a head and shoulders avatar from a provided video of a subject. The main innovation proposed in the approach is to employ regular Gaussian splatting for the face along with sparse anchor Gaussians and a view-dependent neural texture ... | {
"decision": "Accept (Poster)"
} |
HtvZCGiATs | 2402.06223v1 | Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning | {
"content": "## Abstract\n\nAbstract Multimodal contrastive representation learning methods have proven successful across a range of domains, partly due to their ability to generate meaningful shared representations of complex phenomena. To enhance the depth of analysis and understanding of these acquired representa... | [
{
"id": "9K0zJyq7KF",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes a new model for understanding how different types of data (e.g., text and images) relate to each other, moving away from the usual directed acy... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.533735"
} | {
"id": "cFYYFwFhrA",
"metareview": "This paper was about learning a causal model based off of multimodal data with multimodal contrastive learning. The main result identifies variables in the causal model they pose up to transformations. This analysis surfaces the potential for disentanglement for contrastive lear... | {
"decision": "Reject"
} |
Hu0FSOSEyS | 2410.10792v1 | Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations | {
"content": "## Abstract\n\nAbstract Generative models transform random noise into images; their inversion aims to transform images back to structured noise for recovery and editing.\nThis paper addresses two key tasks: (i) inversion and (ii) editing of a real image using stochastic equivalents of rectified flow mod... | [
{
"id": "5rLeD0Ff9s",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper propose RF inversion using dynamic optimal control derived via a linear quadratic regulator. We prove that the resulting vector field is equivalent to ... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;5;4",
"confidence_avg": 4.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;2;3;4",
"contribution_avg": 3,
"presentation": "3;3;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.534833"
} | {
"id": "hoFDRLUBqR",
"metareview": "This paper works on the image inversion and editing by using the stochastic equivalents of rectified flow models. The paper proposed the inversion of rectified flow method by dynamic optimal control via a linear quadratic regulator, and the resulting vector field is a rectified ... | {
"decision": "Accept (Poster)"
} |
HuC8dszO8r | 2405.15421v1 | Model-free reinforcement learning with noisy actions for automated experimental control in optics | {
"content": "## Abstract\n\nAbstract Experimental control involves a lot of manual effort with non-trivial decisions for precise adjustments. Here, we study the automatic experimental alignment for coupling laser light into an optical fiber using reinforcement learning (RL). We face several real-world challenges, su... | [
{
"id": "9SvSrtm0Bt",
"initial_rating": 5,
"confidence": 4,
"soundness": 4,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents an application of deep reinforcement learning algorithms to a physical task of optical fiber coupling. As designed by the authors, the RL agen... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "1;2;4;4",
"soundness_avg": 2.75,
"contribution": "1;1;2;4",
"contribution_avg": 2,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.537272"
} | {
"id": "X69IYM332B",
"metareview": "This paper proposes an application of model-free RL, using soft actor-critic and truncated quantile critics, to automate laser alignment tasks in optical systems. This is primarily an application paper on a novel domain, at least for the ICLR community. They train their agent on... | {
"decision": "Reject"
} |
HuNoNfiQqH | 2406.09289v2 | Understanding Jailbreak Success: A Study of Latent Space Dynamics in Large Language Models | {
"content": "## Abstract\n\nAbstract Conversational large language models are trained to refuse to answer harmful questions. However, emergent jailbreaking techniques can still elicit unsafe outputs, presenting an ongoing challenge for model alignment. To better understand how different jailbreak types circumvent sa... | [
{
"id": "a3Ib3PIiQN",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper explores how jailbreaking techniques work by analyzing latent space dynamics in large language models. The authors found that different types of jailbr... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;3;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.542507"
} | {
"id": "vy6Zz5KYPe",
"metareview": "This paper studying common underlying mechanisms for different types of jailbreaks in language models and how they can be potentially used to counteract jailbreaking. The reviewers agreed that the paper makes an interesting contribution to our understanding of jailbreak success ... | {
"decision": "Reject"
} |
Hx2ADQLi8M | 2410.01481v1 | SonicSim: A customizable simulation platform for speech processing in moving sound source scenarios | {
"content": "## Abstract\n\nAbstract The systematic evaluation of speech separation and enhancement models under moving sound source conditions typically requires extensive data comprising diverse scenarios.\nHowever, real-world datasets often contain insufficient data to meet models’ training and evaluation require... | [
{
"id": "WGbNKCGyES",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes a system and dataset for stimulating moving source sound data. The system is built using Habitat-sim which takes in a 3D scene and stimulates a... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.543244"
} | {
"id": "H010gn1sEh",
"metareview": "This paper introduces SonicSim, a simulation platform built on Habitat-sim that generates synthetic data for speech separation and enhancement in scenarios with moving sound sources. While reviewers recognized the value of such an advanced simulation tool for these dynamic condi... | {
"decision": "Accept (Poster)"
} |
HxKSzulSD1 | 2406.11431v2 | Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization | {
"content": "## Abstract\n\nAbstract Superalignment, where humans act as weak supervisors for superhuman models, has become a crucial problem with the rapid development of Large Language Models (LLMs).\nRecent work has preliminarily studied this problem by using weak models to supervise strong models, and discovered... | [
{
"id": "4BFuDxRxFp",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper explores the phenomenon of weak-to-strong deception, where strong AI models supervised by weaker models can appear aligned in familiar areas while misal... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.544698"
} | {
"id": "t3dnpfUUO3",
"metareview": "This paper investigates the weak-to-strong generalization phenomenon, where strong models supervised by weaker ones surpass their teachers. The authors raise concerns about weak-to-strong deception. This occurs when strong models appear aligned in areas understood by weak models... | {
"decision": "Accept (Poster)"
} |
Hxm0hOxph2 | 2402.04875v4 | On Provable Length and Compositional Generalization | {
"content": "## Abstract\n\nAbstract Out-of-distribution generalization capabilities of sequence-to-sequence models can be studied from the lens of two crucial forms of generalization: length generalization – the ability to generalize to longer sequences than ones seen during training, and compositional generalizati... | [
{
"id": "97oXncXANO",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper addresses two key challenges in out-of-distribution generalization for seq-to-seq models: length generalization and compositional generalization. While... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "3;4;3;4",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.545907"
} | {
"id": "CjNz5gvwZ4",
"metareview": "This submission aims to provide theoretical guarantees on length and compositional generalization for various simplified sequence-to-sequence architectures (deep sets, single-layer transformer variants, simplified RNNs, and SSMs) under the realizability assumption and certain re... | {
"decision": "Reject"
} |
HyPofygOCT | 2312.05821v4 | ASVD: Activation-aware Singular Value Decomposition for Compressing Large Language Models | {
"content": "## Abstract\n\nAbstract In this paper, we introduce a new post-training compression paradigm for Large Language Models (LLMs) to facilitate their wider adoption. We delve into LLM weight low-rank decomposition, and find that the challenges of this task stem from ❶ the distribution variance in the LLM ac... | [
{
"id": "vwBXfcYxgN",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes Activation-aware Singular Value Decomposition (ASVD) to prune the weights of LLMs in a training-free manner. In particular, the technical cont... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "4;5;4;3",
"confidence_avg": 4,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.547379"
} | {
"id": "rBaUaxQfXt",
"metareview": "The paper proposes a novel post-training compression method called Activation-aware Singular Value Decomposition (ASVD) for Large Language Models (LLMs), highlighting its ability to reduce model size efficiently while maintaining performance. This work is significant as it addre... | {
"decision": "Reject"
} |
HzbPIqvhGg | 2409.14011v1 | Generalizable Non-Line-of-Sight Imaging with Learnable Physical Priors | {
"content": "## Abstract\n\nAbstract Non-line-of-sight (NLOS) imaging, recovering the hidden volume from indirect reflections, has attracted increasing attention due to its potential applications. Despite promising results, existing NLOS reconstruction approaches are constrained by the reliance on empirical physical... | [
{
"id": "6akRQC9b2b",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes to learn the physical priors for non-line-of-sight imaging. More specifically two components are proposed, learnable path compensation to atte... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;2;4;4",
"confidence_avg": 3.5,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.548545"
} | {
"id": "8fytQWU1Bp",
"metareview": "This paper aims to enhance non-line-of-sight (NLOS) imaging performance through active time-transient imaging with SPAD. The authors argue that traditional methods rely on empirical physical priors, which lack generalization abilities. They demonstrated improved results with bot... | {
"decision": "Reject"
} |
I1MKOjNVup | 2407.00466v1 | BioKGBench: A Knowledge Graph Checking Benchmark of AI Agent for Biomedical Science | {
"content": "## Abstract\n\nAbstract Pursuing artificial intelligence for biomedical science, a.k.a. AI Scientist, draws increasing attention, where one common approach is to build a copilot agent driven by Large Language Models (LLMs).\nHowever, to evaluate such systems, people either rely on direct Question-Answer... | [
{
"id": "JilLSMbsko",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes several tasks related to the scientific literature review and evaluates the performance of LLMs and LLM-based agents. The authors also propose... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "1;2;2;3",
"soundness_avg": 2,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.549321"
} | {
"id": "LxM8DhFglp",
"metareview": "This paper introduces BioKGBench, a novel benchmark designed to evaluate the capability of LLMs and AI-driven agents in biomedical science. The study distinguishes itself by attempting to address the challenges of factual verification and literature grounding in a systematic way... | {
"decision": "Reject"
} |
I1VCj1l1Zn | 2410.01497v1 | DLP-LoRA: Efficient Task-Specific LoRA Fusion with a Dynamic, Lightweight Plugin for Large Language Models | {
"content": "## Abstract\n\nAbstract Recent advancements in Large Language Models (LLMs) have achieved robust performance across diverse tasks, but fine-tuning these models for specific domains remains resource-intensive. Parameter-Efficient Fine-Tuning (PEFT) methods like Low-Rank Adaptation (LoRA) address this cha... | [
{
"id": "k5SmgUEOkS",
"initial_rating": 3,
"confidence": 3,
"soundness": 1,
"contribution": 1,
"presentation": 2,
"summary": "This paper proposes a method to dynamically fuse pre-trained task-specific LoRA modules. The idea is to train a sentence-level router for different tasks, and sim... | {
"rating": "3;3;3",
"rating_avg": 3,
"confidence": "3;4;3",
"confidence_avg": 3.3333333333333335,
"soundness": "2;2;1",
"soundness_avg": 1.6666666666666667,
"contribution": "2;2;1",
"contribution_avg": 1.6666666666666667,
"presentation": "2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.551640"
} | {
"id": "R78NAh02Lb",
"metareview": "All reviewers agree that the paper addresses an important problem and the methodology looks promising. But there is a consensus among the reviewers that the comparison has been done only with the simple baselines (in terms of performance) and not with the more sophisticated ones... | {
"decision": "Reject"
} |
I393kV3bz4 | 2409.02483v2 | TASAR: Transfer-based Attack on Skeletal Action Recognition | {
"content": "## Abstract\n\nAbstract Skeletal sequences, as well-structured representations of human behaviors, are crucial in Human Activity Recognition (HAR). The transferability of adversarial skeletal sequences enables attacks in real-world HAR scenarios, such as autonomous driving, intelligent surveillance, and... | [
{
"id": "1wr6gFOJQ8",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper studies transfer-based attacks for skeleton-based action recognition. A last-layer Bayesian approches is applied to smooth the loss surface to improve ... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "3;4;3;3",
"soundness_avg": 3.25,
"contribution": "3;3;3;2",
"contribution_avg": 2.75,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.552318"
} | {
"id": "HkpFKfTtbQ",
"metareview": "The paper presents a well-written, easy-to-apply approach for transfer-based attacks for skeleton-based action recognition. The new approach improves adversarial transferability and addresses limitations in S-HAR attacks, demonstrated by thorough experiments and ablation studies... | {
"decision": "Accept (Poster)"
} |
I4YAIwrsXa | 2408.08152v1 | Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search | {
"content": "## Abstract\n\nAbstract We introduce DeepSeek-Prover-V1.5, an open-source language model designed for theorem proving in Lean 4, which enhances DeepSeek-Prover-V1 by optimizing both training and inference processes. Pre-trained on DeepSeekMath-Base with specialization in formal mathematical languages, t... | [
{
"id": "V1sng1mZjx",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents a method for improving LLM performance on theorem proving tasks, using RL on feedback from a proof assistant tool Lean to finetune the model, ... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;3;3",
"contribution_avg": 2.25,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.553059"
} | {
"id": "JlkdXbAgSR",
"metareview": "This paper uses proof assistant feedback to improve the capabilities of LLMs to construct formal proofs in Lean. It casts the problem in the RL paradigm using Lean verification outcomes as the reward signal. The generation of diverse proofs is encouraged by employing a variant o... | {
"decision": "Accept (Poster)"
} |
I4YU0oECtK | 2410.16531v3 | Bayesian scaling laws for in-context learning | {
"content": "## Abstract\n\nAbstract In-context learning (ICL) is a powerful technique for getting language models to perform complex tasks with no training updates.\nPrior work has established strong correlations between the number of in-context examples provided and the accuracy of the model’s predictions.\nIn thi... | [
{
"id": "zhzuZt5Gv2",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper makes progress towards understanding (1) scaling laws for ICL and (2) understanding whether ICL is bayesian. In particular, it presents a scaling law f... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "3;4;2;3",
"confidence_avg": 3,
"soundness": "1;1;2;3",
"soundness_avg": 1.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;4;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.553850"
} | {
"id": "6cYGlqHebe",
"metareview": "This paper proposes a Bayesian scaling law for in-context learning. The method is motivated by theory, and the paper conducts empirical evaluations to verify the fitting ability of the proposed scaling law. The experiments are conducted by firstly suppressing certain capability ... | {
"decision": "Reject"
} |
I4e82CIDxv | 2403.19647v2 | Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models | {
"content": "## Abstract\n\nAbstract We introduce methods for discovering and applying sparse feature circuits . These are causally implicated subnetworks of human-interpretable features for explaining language model behaviors. Circuits identified in prior work consist of polysemantic and difficult-to-interpret unit... | [
{
"id": "MCxyAGPmAm",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces sparse feature circuits as a causal network built from human-interpretable features, rather than neurons. The author discuss how to quantify... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;4;4",
"contribution_avg": 3.5,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.554788"
} | {
"id": "UdCOXOj7J5",
"metareview": "Meta-Review\n\nThis submission presents a method for discovering and editing interpretable causal structures—referred to as sparse feature circuits—within large language models (LLMs). Previous work in mechanistic interpretability has often focused on more coarse units (e.g., en... | {
"decision": "Accept (Oral)"
} |
I6UbnkUveF | 2410.22322v1 | Optimizing Posterior Samples for Bayesian Optimization via Rootfinding | {
"content": "## Abstract\n\nAbstract Bayesian optimization devolves the global optimization of a costly objective function\nto the global optimization of a sequence of acquisition functions.\nThis inner-loop optimization can be catastrophically difficult if it involves posterior samples,\nespecially in higher dimens... | [
{
"id": "UL4GbgPGGw",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes a method for optimizing the sample paths generated by GP posterior. The method, TS-root, globally optimizes the posterior samples via gradient-... | {
"rating": "5;5;8;8",
"rating_avg": 6.5,
"confidence": "4;3;2;4",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.555759"
} | {
"id": "8N2X9cPRI0",
"metareview": "This study proposes a novel method for the global optimization of acquisition functions by optimizing sample paths generated from the posterior distribution of Gaussian processes (GPs). The key feature of this method is its utilization of the decoupled representation of the GP p... | {
"decision": "Accept (Poster)"
} |
I7uCwGxVnl | 2408.02666v2 | Self-Taught Evaluators | {
"content": "## Abstract\n\nAbstract Model-based evaluation is at the heart of successful model\ndevelopment –\nas a reward model for training, and as a replacement for human evaluation.\nTo train such evaluators, the standard approach is to collect a large amount of human preference judgments over model responses, ... | [
{
"id": "KH7hwYFCE9",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper presents a novel approach to training LLMs to act as **evaluators**, or \"judges,\" without relying on human-labeled data. The key method involves a se... | {
"rating": "5;5;5;6;6",
"rating_avg": 5.4,
"confidence": "4;4;3;5;3",
"confidence_avg": 3.8,
"soundness": "2;3;3;3;3",
"soundness_avg": 2.8,
"contribution": "4;3;3;3;3",
"contribution_avg": 3.2,
"presentation": "2;4;2;2;4",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.556498"
} | {
"id": "p9bTAEHpq1",
"metareview": "**summary**\n\nThe paper introduces Self-Taught Evaluator for training LLMs as evaluators without relying on human annotations, using a fully synthetic, iterative self-training process. The method generates contrasting response pairs for given instructions, curating synthetic pr... | {
"decision": "Reject"
} |
I86z54CL2y | 2403.10242v2 | GeoGS3D: Single-view 3D Reconstruction via Geometric-aware Diffusion Model and Gaussian Splatting | {
"content": "## Abstract\n\nAbstract We introduce GeoGS3D, a novel two-stage framework for reconstructing detailed 3D objects from single-view images. Inspired by the success of pre-trained 2D diffusion models, our method incorporates an orthogonal plane decomposition mechanism to extract 3D geometric features from ... | [
{
"id": "EquJGlkED8",
"initial_rating": 3,
"confidence": 2,
"soundness": 1,
"contribution": 2,
"presentation": 2,
"summary": "The manuscript introduces a framework named GeoGS3D, which can reconstruct 3D from a single image. It proposes to perform a decomposed attention mechanism for mul... | {
"rating": "3;3;3;3;5",
"rating_avg": 3.4,
"confidence": "4;4;4;2;4",
"confidence_avg": 3.6,
"soundness": "1;3;3;1;3",
"soundness_avg": 2.2,
"contribution": "1;1;3;2;2",
"contribution_avg": 1.8,
"presentation": "1;2;2;2;1",
"presentation_avg": 1.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.557138"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
I9Dsq0cVo9 | 2410.08942v1 | Maximizing the Potential of Synthetic Data: Insights from Random Matrix Theory | {
"content": "## Abstract\n\nAbstract Synthetic data has gained attention for training large language models, but poor-quality data can harm performance (see, e.g., Shumailov et al. ( 2023 ); Seddik et al. ( 2024 ) ). A potential solution is data pruning, which retains only high-quality data based on a score function... | [
{
"id": "5ERNMOLk7t",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This work considers the problem of maximizing the usefulness of synthetic data in the modern era of generative modelling whereby at least part of the training dat... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;4;5",
"confidence_avg": 4,
"soundness": "3;3;4",
"soundness_avg": 3.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;4",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.558008"
} | {
"id": "pu2lGIuu6M",
"metareview": "The submission considers the impact of using synthetic data (e.g., samples from a pre-trained generative model) to train a model, and how one must incorporate a verification/pruning mechanism. This is investigated first from a theoretical point of view using a Gaussian mixture m... | {
"decision": "Accept (Poster)"
} |
I9bEi6LNgt | 2410.06172v1 | Multimodal Situational Safety | {
"content": "## Abstract\n\nAbstract Multimodal Large Language Models (MLLMs) are rapidly evolving, demonstrating impressive capabilities as multimodal assistants that interact with both humans and their environments. However, this increased sophistication introduces significant safety concerns. In this paper, we pr... | [
{
"id": "PYVUiFWKo0",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 1,
"presentation": 3,
"summary": "The paper presents a novel evaluation framework for assessing the safety of Multimodal Large Language Models in varying situational contexts, termed the Multimoda... | {
"rating": "5;5;6;6;8",
"rating_avg": 6,
"confidence": "4;4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;3;3;3",
"soundness_avg": 2.6,
"contribution": "2;2;3;3;1",
"contribution_avg": 2.2,
"presentation": "3;3;3;4;3",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.559144"
} | {
"id": "cWSnwcg1sG",
"metareview": "The submission is about a novel concept and challenge named \"Multimodal Situational Safety\". The submission initially received mixed ratings. The rebuttal was effective and convinced negative reviewers including KR8B and v6gy. Post rebuttal, all reviewers like the submissio... | {
"decision": "Accept (Poster)"
} |
I9omfcWfMp | 2411.07663v1 | Is Graph Convolution Always Beneficial For Every Feature? | {
"content": "## Abstract\n\nAbstract Graph Neural Networks (GNNs) have demonstrated strong capabilities in processing structured data. While traditional GNNs typically treat each feature dimension equally during graph convolution, we raise an important question: Is the graph convolution operation equally beneficial ... | [
{
"id": "OwW6jWkhPb",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper investigates the problem of feature selection for graph Convolution Networks (GCNs). It begins by introducing TFI, a metric designed to guide the selec... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;5;3;4",
"confidence_avg": 4,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.560008"
} | {
"id": "VoaTV0tL9r",
"metareview": "This paper proposes a novel technique called e Graph Feature Selection (GFS) to enhance the performance of GCNs. The method leverages a Topological Feature Informativeness (TFI) to distinguish GNN-favored features and GNN-disfavored features. The proposed method is justified by ... | {
"decision": "Accept (Poster)"
} |
IC5RJvRoMp | 2403.19135v4 | Streamlining Redundant Layers to Compress Large Language Models | {
"content": "## Abstract\n\nAbstract This paper introduces LLM-Streamline, a pioneer work on layer pruning for large language models (LLMs). It is based on the observation that different layers have varying impacts on hidden states, enabling the identification of less important layers to be pruned.\nLLM-Streamline c... | [
{
"id": "vTRpXaUW9j",
"initial_rating": 8,
"confidence": 1,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This submission presented LLM-Streamline, a new model pruning method for LLMs. Besides traditional pruning, this method also proposed layer replacement, a novel m... | {
"rating": "3;8;8;8",
"rating_avg": 6.75,
"confidence": "4;4;5;1",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;4;3",
"contribution_avg": 3,
"presentation": "2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.561353"
} | {
"id": "SktqUIjpB5",
"metareview": "The authors propose a method for layer pruning and so-called replacement in LLMs to minimise the impact of pruning. They also consider a dedicated metric to asses the resulting model compression. The claims are substantiated experimentally and were considered convincing by all r... | {
"decision": "Accept (Spotlight)"
} |
IDJUscOjM3 | 2406.12034v2 | Self-MoE: Towards Compositional Large Language Models with Self-Specialized Experts | {
"content": "## Abstract\n\nAbstract We present Self-MoE, an approach that transforms a monolithic LLM into a compositional, modular system of self-specialized experts, named MiXSE (MiXture of Self-specialized Experts). Our approach leverages self-specialization, which constructs expert modules using self-generated ... | [
{
"id": "WtIV0zrmoc",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a novel approach to modularize large language models by constructing expert modules from self-generated synthetic data and creating a composit... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.562184"
} | {
"id": "fZnSvIQKGi",
"metareview": "This paper proposes a novel approach to transform monolithic LLMs into a mixture-of-experts system with self-specialized experts using synthetic data. The key scientific claim is that Self-MoE can improve model performance across diverse domains while mitigating catastrophic for... | {
"decision": "Accept (Poster)"
} |
IDxZhXrpNf | 2409.11321v1 | SOAP: Improving and Stabilizing Shampoo using Adam | {
"content": "## Abstract\n\nAbstract There is growing evidence of the effectiveness of Shampoo, a higher-order preconditioning method, over Adam in deep learning optimization tasks. However, Shampoo’s drawbacks include additional hyperparameters and computational overhead when compared to Adam, which only updates ru... | [
{
"id": "apoLy84jrg",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces SOAP (ShampoO with Adam in the Preconditioner’s eigenbasis), an optimizer designed to improve the efficiency and stability of Shampoo by int... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "3;4;3;2",
"confidence_avg": 3,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.563037"
} | {
"id": "v6qizbGyNN",
"metareview": "This paper proposes SOAP, a novel optimization method that builds on the insight that Shampoo is equivalent to running Adafactor in the eigenbasis of Shampoo’s preconditioner. SOAP updates the running average of the second moment in a manner similar to Adam, achieving significan... | {
"decision": "Accept (Poster)"
} |
IEMmEd5Jgm | 2411.01123v1 | X-Drive: Cross-modality Consistent Multi-Sensor Data Synthesis for Driving Scenarios | {
"content": "## Abstract\n\nAbstract Recent advancements have exploited diffusion models for the synthesis of either LiDAR point clouds or camera image data in driving scenarios.\nDespite their success in modeling single-modality data marginal distribution, there is an under-exploration in the mutual reliance betwee... | [
{
"id": "79oRGaS7fJ",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces X-DRIVE, a novel framework for generating consistent multimodal data, specifically LiDAR point clouds and multi-view images, for autonomous ... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.563741"
} | {
"id": "cACfVa86lO",
"metareview": "The paper proposes a method to synthesize consistent lidar and camera images for outdoor driving scenario using diffusion models. While it bares similarities to methods in single modality generation, the authors use an additional branch where the consistency between the two are ... | {
"decision": "Accept (Poster)"
} |
IEs29RYxfK | 2408.17253v2 | VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasters | {
"content": "## Abstract\n\nAbstract Foundation models have emerged as a promising approach in time series forecasting (TSF). Existing approaches either repurpose large language models (LLMs) or build large-scale time series datasets to develop TSF foundation models for universal forecasting. However, these methods ... | [
{
"id": "2P0vhyXhEU",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper titled \"VISIONTS: VISUAL MASKED AUTOENCODERS ARE FREE-LUNCH ZERO-SHOT TIME SERIES FORECASTERS\" introduces a approach to time series forecasting (TSF) ... | {
"rating": "3;5;8",
"rating_avg": 5.333333333333333,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "1;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.564483"
} | {
"id": "DSPaNRc9Ck",
"metareview": "This paper proposes VisionTS, an approach that reveals the intrinsic similarities between time series and images, reformulates time series forecasting into a masked image modeling task and employs ImageNet-pretrained masked autoencoder (MAE) for zero-shot and full-shot forecasti... | {
"decision": "Reject"
} |
IHRQif8VQC | 2408.05446v1 | Ensemble everything everywhere: Multi-scale aggregation for adversarial robustness | {
"content": "## Abstract\n\nAbstract Adversarial examples pose a significant challenge to the robustness, reliability and alignment of deep neural networks. We propose a novel, easy-to-use approach to achieving high-quality representations that lead to adversarial robustness through the use of multi-resolution input... | [
{
"id": "9IAC4KZRUW",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 4,
"presentation": 4,
"summary": "This paper proposes an innovative method for adversarial robustness in deep learning through multi-scale input processing and dynamic self-ensembling. By introduc... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "4;5;4;3",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "4;3;3;4",
"contribution_avg": 3.5,
"presentation": "4;3;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.565206"
} | {
"id": "21k1UHISsK",
"metareview": "This paper proposes a defense against adversarial attacks that uses hierarchical self-ensembles which are augmented using jittering and noise. While the paper has received good initial reviews, the validity of the approach has been publicly challenged during the discussion. In p... | {
"decision": "Reject"
} |
IHp3vOVQO2 | 2410.14582v3 | Do LLMs estimate uncertainty well in instruction-following? | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) could be valuable personal AI agents across various domains, provided they can precisely follow user instructions. However, recent studies have shown significant limitations in LLMs’ instruction-following capabilities, raising concerns about their rel... | [
{
"id": "piOfxfMxWe",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper evaluates how well LLMs can estimate uncertainty when tasked with instruction following, given that existing research largely focuses on uncertainty in ... | {
"rating": "1;5;8;8",
"rating_avg": 5.5,
"confidence": "3;2;3;3",
"confidence_avg": 2.75,
"soundness": "1;3;4;3",
"soundness_avg": 2.75,
"contribution": "1;3;3;3",
"contribution_avg": 2.5,
"presentation": "1;2;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.566069"
} | {
"id": "hnB7WoTmY6",
"metareview": "The paper introduces a controlled benchmark to assess how LLMs estimate their uncertainty when following instructions. All of the reviewers (except one) agree that the paper is above the bar for publication at ICLR. The one dissenting reviewer did not back up their score with co... | {
"decision": "Accept (Poster)"
} |
II81zQUS1x | 2409.17582v2 | Multiplicative Logit Adjustment Approximates Neural-Collapse-Aware Decision Boundary Adjustment | {
"content": "## Abstract\n\nAbstract Real-world data distributions are often highly skewed. This has spurred a growing body of research on long-tailed recognition, aimed at addressing the imbalance in training classification models. Among the methods studied, multiplicative logit adjustment (MLA) stands out as a sim... | [
{
"id": "MiW2i4PoF5",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper provides a theoretical foundation for the Multiplicative Logit Adjustment (MLA) method used in long-tailed recognition tasks. First, the authors develo... | {
"rating": "3;6;6",
"rating_avg": 5,
"confidence": "4;2;3",
"confidence_avg": 3,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.566931"
} | {
"id": "7yEbHFfTfm",
"metareview": "This paper studies the multiplicative logit adjustment (MLA) method proposed recently for the long-tail recognition (LTR) problem. The paper formally shows that MLA can be seen as an approximation of a theoretically-guided approach based on a neural collapse (NC) based adjusting... | {
"decision": "Accept (Poster)"
} |
IIDFStLGQx | 2405.18003v1 | MAVIN: Multi-Action Video Generation with Diffusion Models via Transition Video Infilling | {
"content": "## Abstract\n\nAbstract Diffusion-based video generation has achieved significant progress, yet generating multiple actions that occur sequentially remains a formidable task.\nDirectly generating a video with sequential actions can be extremely challenging due to the scarcity of fine-grained action anno... | [
{
"id": "00H0vH63pZ",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 4,
"summary": "This paper introduces MAVIN, a framework for the transition video infilling task, which aims to generate an intermediate video clip between two adjoining clips, e... | {
"rating": "5;5;5;5;6",
"rating_avg": 5.2,
"confidence": "4;4;2;4;4",
"confidence_avg": 3.6,
"soundness": "3;3;3;2;3",
"soundness_avg": 2.8,
"contribution": "2;2;3;2;3",
"contribution_avg": 2.4,
"presentation": "3;3;3;4;3",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.567796"
} | {
"id": "z7WOex3WII",
"metareview": "Despite the interesting problem context and the proposed MAVIN model, the reviewers' concerns about limited novelty, insufficient experimental validation and lack of clarity. Therefore the paper's contributions are considered insufficient for acceptance on ICLR, and a rejection ... | {
"decision": "Reject"
} |
IIVYiJ1ggK | 2410.06577v1 | Rodimus*: Breaking the Accuracy-Efficiency Trade-Off with Efficient Attentions | {
"content": "## Abstract\n\nAbstract Recent advancements in Transformer-based large language models (LLMs) have set new standards in natural language processing. However, the classical softmax attention incurs significant computational costs, leading to a 𝒪 ( T ) 𝒪 𝑇 {\\mathcal{O}}(T) caligraphic_O ( italic_T )... | [
{
"id": "1lDZOHOWhe",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces two efficient Transformer architectures that incorporate semantic, token, and head compression: Rodimus, which is purely based on linear rec... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;5;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;2",
"contribution_avg": 2.5,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.568608"
} | {
"id": "cfmrIO1tpE",
"metareview": "Based on the consistent reviews, I recommend accepting this paper. While some reviewers note incremental aspects and missing efficiency metrics, the novel contributions of Rodimus and Rodimus+ demonstrate significant improvements in the accuracy-efficiency trade-off for large la... | {
"decision": "Accept (Poster)"
} |
IIzehISTBe | 2410.06703v2 | ST-WebAgentBench: A Benchmark for Evaluating Safety and Trustworthiness in Web Agents | {
"content": "## Abstract\n\nAbstract Recent advancements in LLM-based web agents have introduced novel architectures and benchmarks showcasing progress in autonomous web navigation and interaction. However, most existing benchmarks prioritize effectiveness and accuracy, overlooking crucial factors like safety and tr... | [
{
"id": "BA4Wri38Cu",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper presents a benchmark for evaluating \"safety and trustworthiness in web agents\" - where safety and trustworthiness are defined based on 12 dimensions ... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;1;2;3",
"soundness_avg": 2.25,
"contribution": "3;1;3;3",
"contribution_avg": 2.5,
"presentation": "2;3;2;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.569775"
} | {
"id": "2GB3DMXFrj",
"metareview": "This paper introduces a benchmark for assessing “safety and trustworthiness in web agents,” defined across 12 key dimensions. The benchmark contains 235 policy-enriched tasks. Each task specifies an action goal (e.g., creating a GitLab user group) and includes policies requiring... | {
"decision": "Reject"
} |
IJiTI0fB0e | 2411.02817v1 | An Information-Theoretic Approach to Diversity Evaluation of Prompt-based Generative Models | {
"content": "## Abstract\n\nAbstract Text-conditioned generation models are commonly evaluated based on the quality of the generated data and its alignment with the input text prompt. On the other hand, several applications of prompt-based generative models require sufficient diversity in the generated data to ensur... | [
{
"id": "tjked64aPM",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This paper proposes a metric for evaluating the diversity of generative models. Unlike previous approaches which focus on unconditional generations, this paper us... | {
"rating": "3;3;3;8",
"rating_avg": 4.25,
"confidence": "3;4;5;3",
"confidence_avg": 3.75,
"soundness": "3;3;2;4",
"soundness_avg": 3,
"contribution": "1;2;1;3",
"contribution_avg": 1.75,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.570467"
} | {
"id": "kJxg9IWJF8",
"metareview": "This work presents an information-theoretic approach to evaluate the diversity of prompt-based content generation. The proposed evaluation score defined in Section 4 consists of two parts: Conditional-Vendi and Information-Vendi. A statistical interpretation of the proposed scor... | {
"decision": "Reject"
} |
IK7l0CqZuH | 2408.08201v1 | Heavy Labels Out! Dataset Distillation with Label Space Lightening | {
"content": "## Abstract\n\nAbstract Dataset distillation or condensation aims to condense a large-scale training dataset into a much smaller synthetic one such that the training performance of distilled and original sets on neural networks are similar. Although the number of training samples can be reduced substant... | [
{
"id": "4QyuCfpoGz",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper addresses the high storage cost of soft labels in dataset distillation that generates synthetic labels directly from images. It leverages open-source m... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;5;3;3",
"confidence_avg": 3.75,
"soundness": "2;3;1;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;2;1;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.571227"
} | {
"id": "3spqLNpMEQ",
"metareview": "The paper introduces an innovative method to tackle the high storage costs related to soft labels in dataset distillation by employing open-source models and a fine-tuning strategy to derive labels directly from images. However, the primary concerns revolve around the reliance o... | {
"decision": "Reject"
} |
IL85Ebjg9j | 2406.00958v1 | Navigating Conflicting Views: Harnessing Trust for Learning | {
"content": "## Abstract\n\nAbstract Resolving conflicts is essential to make the decisions of multi-view classification more reliable.\nMuch research has been conducted on learning consistent informative representations among different views, assuming that all views are identically important and strictly aligned.\n... | [
{
"id": "8nclTYSqAp",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "Not all views can be trusted to the same degree in the field of Multi-View Classification. Trust Discounting-Based Trust Fusion (TDTF) is proposed to model comput... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "5;4;3;2",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.572137"
} | {
"id": "YbSuEIfOCJ",
"metareview": "The authors study learning from multiple views. The technical contribution uses an existing fusion method [1] to deal with conflicting views which has also been explored in [2] but it does not really become obvious what the advantage/difference over [2] is. Also the empirical ev... | {
"decision": "Reject"
} |
ILSZZNlbqw | 2406.01899v2 | Cross-Domain Graph Data Scaling: A Showcase with Diffusion Models | {
"content": "## Abstract\n\nAbstract Models for natural language and images benefit from data scaling behavior: the more data fed into the model, the better they perform.\nThis ’better with more’ phenomenon enables the effectiveness of large-scale pre-training on vast amounts of data.\nHowever, current graph pre-tra... | [
{
"id": "dSrfTWWJIi",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes a graph structure augmentation framework built on discrete diffusion models. They first pre-train a structure-only discrete diffusion model on ... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;4;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.66666666666666... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.572949"
} | {
"id": "nNgswQ7yiX",
"metareview": "The UniAug framework proposed in this paper shows some potential in cross-domain graph learning tasks, but several key issues remain. First, there is inconsistency in the comparison with baseline methods, particularly regarding the differences in feature usage, which affects the... | {
"decision": "Reject"
} |
INFfvQArFY | 2410.06331v2 | Locate-then-edit for Multi-hop Factual Recall under Knowledge Editing | {
"content": "## Abstract\n\nAbstract The locate-then-edit paradigm has shown significant promise for knowledge editing (KE) in Large Language Models (LLMs). While previous methods perform well on single-hop fact recall tasks, they consistently struggle with multi-hop factual recall tasks involving newly edited knowl... | [
{
"id": "wXMb59y6tU",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper addresses challenges in existing locate-then-edit knowledge editing methods, particularly focusing on multi-hop knowledge editing. The authors first inv... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;3;2;4",
"soundness_avg": 3,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "3;1;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.573688"
} | {
"id": "39EYD57t66",
"metareview": "This paper examines multi-hop knowledge editing in large language models, highlighting the shortcomings of current locate-then-edit methods. The authors' mechanistic interpretability analysis reveals that multi-hop knowledge queries process information differently from single-ho... | {
"decision": "Reject"
} |
INe4otjryz | 2410.09695v2 | Can In-context Learning Really Generalize to Out-of-distribution Tasks? | {
"content": "## Abstract\n\nAbstract In this work, we explore the mechanism of in-context learning (ICL) on out-of-distribution (OOD) tasks that were not encountered during training. To achieve this, we conduct synthetic experiments where the objective is to learn OOD mathematical functions through ICL using a GPT-2... | [
{
"id": "0OfL46c2dW",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper studied the limitations of in-context learning (ICL) on out-of-distribution (OOD) tasks through empirical experiments and theoretical analysis. They po... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "3;2;2;4",
"soundness_avg": 2.75,
"contribution": "1;2;2;4",
"contribution_avg": 2.25,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.574377"
} | {
"id": "LTf4ZvtjHO",
"metareview": "This paper investigates the behavior of in-context learning (ICL) on out-of-distribution tasks through empirical analysis and theory development. The reviewers generally found merit in the paper's experimental methodology and results, particularly highlighting the well-designed ... | {
"decision": "Accept (Poster)"
} |
IQ0BBfbYR2 | 2406.01649v1 | CoLa-DCE – Concept-guided Latent Diffusion Counterfactual Explanations | {
"content": "## Abstract\n\nAbstract Recent advancements in generative AI have introduced novel prospects and practical implementations. Especially diffusion models show their strength in generating diverse and, at the same time, realistic features, positioning them well for generating counterfactual explanations fo... | [
{
"id": "dd5cDcWdLd",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The author proposes a novel method, CoLa-DCE (Concept-guided Latent Diffusion Counterfactual Explanations), to generate controlled counterfactual explanations for... | {
"rating": "3;3;5",
"rating_avg": 3.6666666666666665,
"confidence": "4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "2;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.575033"
} | {
"id": "y1GmOsMxTd",
"metareview": "The paper proposes CoLa-DCE (Concept-guided Latent Diffusion Counterfactual Explanations), a novel method that generates controlled counterfactual explanations for image classification models using latent diffusion models. The method introduces concept-based spatial conditioning... | {
"decision": "Reject"
} |
IQCwmB63Fd | 2409.06338v1 | Retrieval Or Holistic Understanding? Dolce: Differentiate Our Long Context Evaluation Tasks | {
"content": "## Abstract\n\nAbstract We argue that there are two major distinct capabilities in long context understanding: retrieval and holistic understanding. Understanding and further improving LLMs’ long context capabilities would not be possible without knowing the tasks’ focus categories. We aim to automatica... | [
{
"id": "618MDmFGAj",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces a framework for understanding and improving long-context capabilities in large language models (LLMs) by distinguishing between two key skil... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "2;2;1;3",
"confidence_avg": 2,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "3;2;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.575943"
} | {
"id": "Sz98Gyzgiu",
"metareview": "This paper claims that there are two major distinct capabilities in long context understanding, namely retrieval and holistic understanding. Specifically, this paper presents the Dolce framework, which parameterizes each problem by complexity and redundancy and assigns to one of... | {
"decision": "Reject"
} |
IQafqgqDzF | 2410.04328v1 | OD-Stega: LLM-Based Near-Imperceptible Steganography via Optimized Distributions | {
"content": "## Abstract\n\nAbstract We consider coverless steganography where a Large Language Model (LLM) drives an arithmetic coding decoder to generate stego-texts. An efficient method should embed secret message bits in as few language tokens as possible, while still keeping the stego-text natural and fluent. W... | [
{
"id": "lczoWQfMa7",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper suggest to distort little bit the probability distribution of LLM in generative steganography to increase the capacity.",
"strengths": "* The theory... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;4;5;3",
"confidence_avg": 4,
"soundness": "3;2;2;2",
"soundness_avg": 2.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "4;3;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.576979"
} | {
"id": "Nsw8ziVCCU",
"metareview": "This paper proposes to distort the probability distribution of LLM in generative steganography. In particular, the distortion of the probability distribution is done by integrating probability truncation and optimized distributions. This approach aims to embed more secret messag... | {
"decision": "Reject"
} |
IRL9wUiwab | 2409.12915v2 | Exploring Representations and Interventions in Time Series Foundation Models | {
"content": "## Abstract\n\nAbstract Time series foundation models (TSFMs) promise to be powerful tools for a wide range of applications. However, their internal representations and learned concepts are still not well understood. In this study, we investigate the structure and redundancy of representations across va... | [
{
"id": "Mx3PyXPFI1",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper aims to understand the internal representations of time series foundation models (TSFM). By analyzing the self-similarity of the model layers, this pap... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "3;3;4",
"confidence_avg": 3.3333333333333335,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.577850"
} | {
"id": "gETpxbvF8P",
"metareview": "This work examines the internal representations of Time Series Foundation Models (TSFMs), identifying block-like redundancy within these representations. By exploiting this redundancy, the authors propose an informed pruning strategy that enhances inference speed and efficiency ... | {
"decision": "Reject"
} |
ISqx8giekS | 2407.10032v2 | LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have shown immense potential across various domains, but their high memory requirements and inference costs remain critical challenges for deployment. Post-training quantization (PTQ) has emerged as a promising technique to reduce memory requirements ... | [
{
"id": "mt45uY3fy6",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a method for optimizing the quantization grid in a layerwise loss-aware manner to reduce low-bit weight quantization error. Unlike previous op... | {
"rating": "3;5;5;5;5;6",
"rating_avg": 4.833333333333333,
"confidence": "3;3;4;4;4;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;3;2;3;3;4",
"soundness_avg": 2.8333333333333335,
"contribution": "2;3;2;3;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "2;2;3;2;3;3",
"prese... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.578590"
} | {
"id": "YO7irY47iB",
"metareview": "This paper proposes LeanQuant, a post-training quantization (PTQ) method to mitigate the memory and inference cost challenges in LLMs. One of the main contributions of the paper is the introduction of loss-error quantization grids, which adapt to outliers in inverse Hessian diag... | {
"decision": "Accept (Poster)"
} |
IT33VLRJuS | 2410.01853v1 | Recovering Time-Varying Networks From Single-Cell Data | {
"content": "## Abstract\n\nAbstract Gene regulation is a dynamic process that underlies all aspects of human development, disease response, and other key biological processes. The reconstruction of temporal gene regulatory networks has conventionally relied on regression analysis, graphical models, or other types o... | [
{
"id": "93jmSu3c7U",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper aims to predict a series of graphs describing gene expression regulation by transcription factors from a series of single-cell gene expression data. The... | {
"rating": "3;3;3;8",
"rating_avg": 4.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;2;2;4",
"soundness_avg": 2.75,
"contribution": "3;2;2;4",
"contribution_avg": 2.75,
"presentation": "2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.579336"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
ITi9Zwkge2 | 2410.02179v1 | HATFormer: Historic Handwritten Arabic Text Recognition with Transformers | {
"content": "## Abstract\n\nAbstract Arabic handwritten text recognition (HTR) is challenging, especially for historical texts, due to diverse writing styles and the intrinsic features of Arabic script.\nAdditionally, Arabic handwriting datasets are smaller compared to English ones, making it difficult to train gene... | [
{
"id": "ILPIN1CHiw",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents an improvement on the transformer-based encoder-decoder architecture of the TrOCR OCR model. These improvements are designed to address the ch... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;2;5;4",
"confidence_avg": 3.75,
"soundness": "2;3;2;2",
"soundness_avg": 2.25,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "3;3;2;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.580108"
} | {
"id": "UkqF9XnLct",
"metareview": "The paper introduces an approach using line warping techniques to adapt elongated rectangular text lines into a square format suitable for ViTs, addressing challenges posed by the non-uniformity of handwritten text. It benefits from extensive pre-training on a synthetic Arabic d... | {
"decision": "Reject"
} |
IUzQfdkkoL | 2409.01966v1 | MetaFood3D: 3D Food Dataset with Nutrition Values | {
"content": "## Abstract\n\nAbstract Food computing is both important and challenging in computer vision (CV). It significantly contributes to the development of CV algorithms due to its frequent presence in datasets across various applications, ranging from classification and instance segmentation to 3D reconstruct... | [
{
"id": "6hIwrxU8BE",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work presents MetaFood3D, a dataset of 3D scans of real-world food items. The dataset consists of about 600 food samples from 108 categories, with multi-view... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "2;4;3;4",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.581150"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
IXGHSVBBCF | 2410.14763v1 | Enabling Scalable Evaluation of Bias Patterns in Medical LLMs | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have shown impressive potential in helping with numerous medical challenges. Deploying LLMs in high-stakes applications such as medicine, however, brings in many concerns. One major area of concern relates to biased behaviors of LLMs in medical applic... | [
{
"id": "nCXXTeOT3t",
"initial_rating": 5,
"confidence": 2,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper presents a method to evaluate bias in medical LLMs by generating clinical vignettes that force LLMs to respond across demographic groups. (Ideally havin... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "3;4;3;2",
"confidence_avg": 3,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "2;3;2;2",
"contribution_avg": 2.25,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.581782"
} | {
"id": "VUiz131djH",
"metareview": "The paper presents a framework for generating clinical vignettes, motivated by scalable bias evaluations. In several medical settings, the paper examine the quality of these generated vignettes including across faithfulness, domain specificity, and bias under counterfactual fair... | {
"decision": "Reject"
} |
IXyfbaGlps | 2406.09588v3 | Learning Color Equivariant Representations | {
"content": "## Abstract\n\nAbstract In this paper, we introduce group convolutional neural networks (GCNNs) equivariant to color variation.\nGCNNs have been designed for a variety of geometric transformations from 2D and 3D rotation groups, to semi-groups such as scale.\nDespite the improved interpretability, accur... | [
{
"id": "omODcYV7fl",
"initial_rating": 5,
"confidence": 2,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces group convolutional neural networks designed to be equivariant to variations in hue, saturation, and color in RGB images. It proposes a lift... | {
"rating": "5;5;5;6;8",
"rating_avg": 5.8,
"confidence": "3;3;2;3;4",
"confidence_avg": 3,
"soundness": "3;4;2;3;4",
"soundness_avg": 3.2,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.582450"
} | {
"id": "AlVH3YizJv",
"metareview": "This submission introduces a colour-equivariant group convolutional network (GCNN) that generalises equivariance beyond the hue channel to saturation in RGB images. The proposed method employs a lifting layer capable of directly transforming input images and further leverages th... | {
"decision": "Accept (Poster)"
} |
IZbthMfqad | 2410.02199v1 | Deep Koopman-layered Model with Universal Property Based on Toeplitz Matrices | {
"content": "## Abstract\n\nAbstract We propose deep Koopman-layered models with learnable parameters in the form of Toeplitz matrices for analyzing the dynamics of time-series data.\nThe proposed model has both theoretical solidness and flexibility.\nBy virtue of the universal property of Toeplitz\nmatrices and the... | [
{
"id": "ZuGSOqY0s8",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper presents a novel deep Koopman-layered framework for modeling dynamical systems, particularly suited for nonautonomous time-series data. This approach in... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "5;2;4;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;1;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.583133"
} | {
"id": "kWxpcyihBE",
"metareview": "This paper proposes to compose a series of learnable Koopman operator approximations to construct \"deep Koopman-layered models\", for the purpose of approximating nonlinear, nonautonomous differential equations. Each Koopman operator (i.e. each \"layer\") has its generator repr... | {
"decision": "Reject"
} |
IZjBfdVRB0 | 2407.19342v2 | Parameter-Efficient Fine-Tuning via Circular Convolution | {
"content": "## Abstract\n\nAbstract Low-Rank Adaptation (LoRA) has gained popularity for fine-tuning large foundation models, leveraging low-rank matrices 𝐀 𝐀 \\mathbf{A} bold_A and 𝐁 𝐁 \\mathbf{B} bold_B to represent weight changes ( i.e., Δ 𝐖 = 𝐁𝐀 Δ 𝐖 𝐁𝐀 \\Delta\\mathbf{W}=\\mathbf{B}\\mathbf{A} roman... | [
{
"id": "hCgkdBat29",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work introduces Circular Convolution Adaptation (C3A) as an advanced parameter-efficient fine-tuning (PEFT) technique, addressing limitations of existing met... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "1;2;3",
"contribution_avg": 2,
"presentation": "4;3;3",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.583911"
} | {
"id": "2iFwRZ79wQ",
"metareview": "(a) Summary\n\nThe paper proposes to use circular convolution as an alternative to the low-rank AB matrices in LoRA/PEFT. The method is called Circular Convolution Adaptation (C3A). It maintains higher rank than LoRA, with more efficiency benefits. It also outperforms LoRA varia... | {
"decision": "Reject"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.