paper_id string | arxiv_id string | title string | markdown dict | reviews list | scores dict | metadata dict | meta_review dict | decision dict |
|---|---|---|---|---|---|---|---|---|
7wuJMvK639 | 2405.18418v2 | Hierarchical World Models as Visual Whole-Body Humanoid Controllers | {
"content": "## Abstract\n\nAbstract Whole-body control for humanoids is challenging due to the high-dimensional nature of the problem, coupled with the inherent instability of a bipedal morphology. Learning from visual observations further exacerbates this difficulty. In this work, we explore highly data-driven app... | [
{
"id": "AKvZS3Lu4W",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes a hierarchical world model for whole-body humanoid control based on RL. The framework separates high-level and low-level control, with a high-l... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.982300"
} | {
"id": "O6g5hLzmvK",
"metareview": "The paper proposes Puppeteer, a hierarchical reinforcement learning (RL)-based framework for whole-body control of humanoid robots based upon visual observations. The high-level model generates reference commands for a pre-trained low-level policy responsible for trajectory trac... | {
"decision": "Accept (Poster)"
} |
7xCSK9BLPy | 2410.02902v3 | Better Instruction-Following Through Minimum Bayes Risk | {
"content": "## Abstract\n\nAbstract General-purpose LLM judges capable of human-level evaluation provide not only a scalable and accurate way of evaluating instruction-following LLMs but also new avenues for supervising and improving their performance. One promising way of leveraging LLM judges for supervision is t... | [
{
"id": "nDuN4zaxOj",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper proposes an approach for selecting one of N generation hypotheses using a Minimum Bayes Risk (MBR) method. MBR decoding alone results in improved perfo... | {
"rating": "5;8;8",
"rating_avg": 7,
"confidence": "4;3;3",
"confidence_avg": 3.3333333333333335,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "1;4;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.983046"
} | {
"id": "6HBJWm5gXx",
"metareview": "This paper introduces a novel application of Minimum Bayes Risk (MBR) decoding to enhance the test-time performance of instruction-following LLMs. LLM judges (including Prometheus2, Llama3, and JudgeLM) are used as reference-based evaluators and select high-quality outputs from ... | {
"decision": "Accept (Spotlight)"
} |
7yncrX80CN | 2404.05825v1 | LLM-Augmented Retrieval: Enhancing Retrieval Models Through Language Models and Doc-Level Embedding | {
"content": "## Abstract\n\nAbstract Recently embedding-based retrieval or dense retrieval have shown state of the art results, compared with traditional sparse or bag-of-words based approaches. This paper introduces a model-agnostic doc-level embedding framework through large language model (LLM) augmentation. In a... | [
{
"id": "UQ6t5xLBtV",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a novel, model-agnostic approach for document embedding, leveraging an augmented document field comprising synthetically generated queries, ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;3;2;2",
"soundness_avg": 2.25,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"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": "2026-01-14T22:16:00.984011"
} | {
"id": "Hxel064AjY",
"metareview": "This paper proposes an LLM-based method for document expansion in document retrieval. It enhances each document with synthetic queries, titles, and chunks generated by the LLM. The proposed method is training-free and model-agnostic because of the use of LLM. The method demonstr... | {
"decision": "Reject"
} |
7zPd1TjRc1 | 2312.17369v1 | SANIA: Polyak-type Optimization Framework Leads to Scale Invariant Stochastic Algorithms | {
"content": "## Abstract\n\nAbstract Adaptive optimization methods are widely recognized as among the most popular approaches for training Deep Neural Networks (DNNs). Techniques such as Adam, AdaGrad, and AdaHessian utilize a preconditioner that modifies the search direction by incorporating information about the c... | [
{
"id": "1DQNFtI4aL",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a training algorithm generalizing the \"Polyak step-size\" to second-order algorithms, such as cubic Newton or quasi-Newton. \n\nLet $f$ be a ... | {
"rating": "6;6;6;6",
"rating_avg": 6,
"confidence": "3;5;3;4",
"confidence_avg": 3.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"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": "2026-01-14T22:16:00.984780"
} | {
"id": "0PQJ773L9F",
"metareview": "This paper studies a general framework for Preconditioned and Second-order Polyak methods. In particular, the paper proposes a first Stochastic Cubic Newton method with polyak step-size and also introduces the new scale invariant versions of AdaGrad and Adam, which make them inv... | {
"decision": "Reject"
} |
81cta3WQVI | 2409.18114v1 | EdgeRunner: Auto-regressive Auto-encoder for Artistic Mesh Generation | {
"content": "## Abstract\n\nAbstract Current auto-regressive mesh generation methods suffer from issues such as incompleteness, insufficient detail, and poor generalization.\nIn this paper, we propose an Auto-regressive Auto-encoder (ArAE) model capable of generating high-quality 3D meshes with up to 4,000 faces at ... | [
{
"id": "cxubEGdgQH",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces EdgeRunner, an Auto-regressive Auto-encoder (ArAE) model for artistic mesh generation. It proposes a mesh tokenization algorithm to address i... | {
"rating": "3;6;6;8;8;8",
"rating_avg": 6.5,
"confidence": "4;4;4;5;3;4",
"confidence_avg": 4,
"soundness": "2;4;3;4;3;4",
"soundness_avg": 3.3333333333333335,
"contribution": "2;4;3;4;3;3",
"contribution_avg": 3.1666666666666665,
"presentation": "3;3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.985494"
} | {
"id": "555qBuaxFI",
"metareview": "The paper introduces EdgeRunner, an auto-regressive auto-encoder with a novel mesh tokenization algorithm for high-quality artistic mesh generation, demonstrating significant advancements in mesh compression and fixed-length latent code learning. Reviewers praised the paper's te... | {
"decision": "Accept (Poster)"
} |
82VzAtBZGk | 2405.18180v1 | Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding | {
"content": "## Abstract\n\nAbstract Empowering safe exploration of reinforcement learning (RL) agents during training is a critical impediment towards deploying RL agents in many real-world scenarios. Training RL agents in unknown, black-box environments poses an even greater safety risk when prior knowledge of the... | [
{
"id": "r28J9iG1WA",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper develops a new shield based on contrastive learning to increase the safety of RL agents. This approach learns a latent representation that separates sa... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;2;3;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.986241"
} | {
"id": "GSDmxWLyvK",
"metareview": "The authors propose a new shielding method based on contrastive learning for safe RL, which learns a latent representation to distinguish between safe and unsafe state-actions. The reviewers all agree to reject the paper, due to weak theoretical results (e.g. unrealistic require... | {
"decision": "Reject"
} |
84pDoCD4lH | 2410.17385v1 | Do Vision-Language Models Represent Space and How? Evaluating Spatial Frame of Reference under Ambiguities | {
"content": "## Abstract\n\nAbstract Spatial expressions in situated communication can be ambiguous, as their meanings vary depending on the frames of reference (FoR) adopted by speakers and listeners.\nWhile spatial language understanding and reasoning by vision-language models (VLMs) have gained increasing attenti... | [
{
"id": "7gBimuieyN",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "The authors assess how VLMs represent space through the lens of “situated communication,” framing the meaning of spatial relations like “to the right of” under tr... | {
"rating": "5;6;8;8;8",
"rating_avg": 7,
"confidence": "3;3;4;4;4",
"confidence_avg": 3.6,
"soundness": "3;4;4;4;4",
"soundness_avg": 3.8,
"contribution": "3;3;3;3;3",
"contribution_avg": 3,
"presentation": "3;3;3;4;4",
"presentation_avg": 3.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.987348"
} | {
"id": "2gPJvm5F5R",
"metareview": "This paper proposes a novel evaluation protocol and benchmark, COnsistent Multilingual Frame Of Reference Test (COMFORT), to examine the spatial reasoning capabilities of vision and language models and evaluates 9 state-of-the-art models using this framework and demonstrate thei... | {
"decision": "Accept (Oral)"
} |
85VWxAwsaF | 2407.17907v1 | Amortized Posterior Sampling with Diffusion Prior Distillation | {
"content": "## Abstract\n\nAbstract We propose a variational inference approach to sample from the posterior distribution for solving inverse problems. From a pre-trained diffusion model, our approach trains a conditional flow model to minimize the divergence between the proposal variational distribution and the po... | [
{
"id": "MJ6FkOocG1",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes to accelerate inverse problem-solving with diffusion priors. This is achieved by distilling the unconditional diffusion prior for a given set o... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;3;5;4",
"confidence_avg": 4,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;1;2",
"contribution_avg": 1.75,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.987997"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
868masI331 | 2410.04380v1 | HALL-E: Hierarchical Neural Codec Language Model for Minute-Long Zero-Shot Text-to-Speech Synthesis | {
"content": "## Abstract\n\nAbstract Recently, Text-to-speech (TTS) models based on large language models (LLMs) that translate natural language text into sequences of discrete audio tokens have gained great research attention, with advances in neural audio codec (NAC) models using residual vector quantization (RVQ)... | [
{
"id": "fMYhlGs3rY",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 1,
"summary": "The paper presents HALL-E, a neural codec language model aimed at addressing challenges in minute-long zero-shot text-to-speech (TTS) synthesis. It introduces Mul... | {
"rating": "5;6;6;6;6",
"rating_avg": 5.8,
"confidence": "4;5;4;4;3",
"confidence_avg": 4,
"soundness": "2;3;3;3;2",
"soundness_avg": 2.6,
"contribution": "4;3;3;4;3",
"contribution_avg": 3.4,
"presentation": "3;3;3;3;1",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.988793"
} | {
"id": "P5bEZETz6r",
"metareview": "In this work, the authors targeted to resolve the challenge of long-form TTS. The high frame rate results in the long length of audio tokens, which makes it different for autoregressive language models to generate tokens. Representative works such as VALL-E encounter challenges ... | {
"decision": "Accept (Poster)"
} |
86hNGGo1CU | 2410.11165v3 | Toward Efficient Kernel-Based Solvers for Nonlinear PDEs | {
"content": "## Abstract\n\nAbstract This paper introduces a novel kernel learning framework toward efficiently solving nonlinear partial differential equations (PDEs). In contrast to the state-of-the-art kernel solver that embeds differential operators within kernels, posing challenges with a large number of colloc... | [
{
"id": "VtIS9wUzU2",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 4,
"presentation": 4,
"summary": "The paper introduces and studies a novel kernel-based method for the approximation of the solution of a broad class of non-linear PDEs. The authors discuss comput... | {
"rating": "1;5;5;5",
"rating_avg": 4,
"confidence": "4;3;3;5",
"confidence_avg": 3.75,
"soundness": "2;3;2;2",
"soundness_avg": 2.25,
"contribution": "1;2;2;4",
"contribution_avg": 2.25,
"presentation": "2;4;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.989611"
} | {
"id": "0kEpRFRmrA",
"metareview": "The paper introduces and studies a novel kernel-based method for the approximation of the solution of a broad class of non-linear PDEs. The authors discuss computational aspects and provide error estimates in Sobolev spaces of appropriate regularity. Unfortunately, two reviewers... | {
"decision": "Reject"
} |
88AS5MQnmC | 2409.13156v1 | RRM: Robust Reward Model Training Mitigates Reward Hacking | {
"content": "## Abstract\n\nAbstract Reward models (RMs) play a pivotal role in aligning large language models (LLMs) with human preferences. However, traditional RM training, which relies on response pairs tied to specific prompts, struggles to disentangle prompt-driven preferences from prompt-independent artifacts... | [
{
"id": "rO13CsdDzS",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "Traditional reward model (RM) training for large language models (LLMs) struggles to separate prompt-driven preferences from irrelevant artifacts like response le... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "3;5;4;4",
"confidence_avg": 4,
"soundness": "1;3;2;4",
"soundness_avg": 2.5,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.990296"
} | {
"id": "wjBLBiC90K",
"metareview": "This paper proposes a robust reward model (RRM) training method to mitigate reward hacking by leveraging a causal inference framework. A causal graph for human preference modeling is introduced to help the model distinguish between contextual preference signals and context-free ... | {
"decision": "Accept (Poster)"
} |
88Qm4fGWzX | 2410.02483v1 | Event-Customized Image Generation | {
"content": "## Abstract\n\nAbstract Customized Image Generation, generating customized images with user-specified concepts, has raised significant attention due to its creativity and novelty. With impressive progress achieved in subject customization, some pioneer works further explored the customization of action ... | [
{
"id": "mazqPrdh9m",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a new task called Event-Customized Image Generation, which aims to not only control subjects, but also customize all specific actions, poses, ... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"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": "2026-01-14T22:16:00.990906"
} | {
"id": "jZkeb2eqf3",
"metareview": "This work obtains two positive and two negative scores. After checking the paper, the AC is still concerned about the limited technical novelty of the proposed application. \n\nAs the author said, \"While these observations have been widely recognized in previous works, we are t... | {
"decision": "Reject"
} |
8CKgS18uWx | 2410.03553v2 | Structure-Enhanced Protein Instruction Tuning: Towards General-Purpose Protein Understanding | {
"content": "## Abstract\n\nAbstract Proteins, as essential biomolecules, play a central role in biological processes, including metabolic reactions and DNA replication. Accurate prediction of their properties and functions is crucial in biological applications. Recent development of protein language models (pLMs) w... | [
{
"id": "kSt05hjtGS",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "The paper introduces structure-enhanced protein instruction tuning (SEPIT) framework to learn a general-purpose protein understanding model. The paper combines a ... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;4;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.991682"
} | {
"id": "r5hnoBJNdp",
"metareview": "The paper introduces the Structure-Enhanced Protein Instruction Tuning (SEPIT) framework, which aims to achieve general-purpose protein understanding by integrating structural knowledge into protein language models and connecting them to large language models. The authors propos... | {
"decision": "Reject"
} |
8DBTq09LgN | 2405.16450v2 | Synthesizing Programmatic Reinforcement Learning Policies with Large Language Model Guided Search | {
"content": "## Abstract\n\nAbstract Programmatic reinforcement learning (PRL) has been explored for representing policies through programs as a means to achieve interpretability and generalization. Despite promising outcomes, current state-of-the-art PRL methods are hindered by sample inefficiency, necessitating te... | [
{
"id": "5V58qJFKAp",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper studies learning programmatic-actions for RL, through heuristic search methods such as hill climbing. The paper focuses on the initialization problem. ... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "4;4;5",
"confidence_avg": 4.333333333333333,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "3;2;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": "2026-01-14T22:16:00.992508"
} | {
"id": "klq6zUdVGR",
"metareview": "The paper presents a mechanism to use LLMs to find policies in a Domain Specific Language (DSL) for the Programmatic Reinforcement Learning paradigm. The technique builds on the Hill Climbing (HC) algorithm that effectively searches for programmatic policies directly in the prog... | {
"decision": "Accept (Poster)"
} |
8Dj6OEMj6W | 2410.10735v1 | Embedding Self-Correction as an Inherent Ability in Large Language Models for Enhanced Mathematical Reasoning | {
"content": "## Abstract\n\nAbstract Accurate mathematical reasoning with Large Language Models (LLMs) is crucial in revolutionizing domains that heavily rely on such reasoning. However, LLMs often encounter difficulties in certain aspects of mathematical reasoning, leading to flawed reasoning and erroneous results.... | [
{
"id": "axANXjzcr8",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents a fine-tuning technique called Chain of Self-correction (CoSC) that embeds self-correction as an inherent ability of LLM. The method is specifi... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"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": "2026-01-14T22:16:00.993145"
} | {
"id": "i8G8NgRMMD",
"metareview": "This paper introduces Chain of Self-Correction (CoSC), a fine-tuning mechanism enabling LLMs to iteratively self-correct by generating programs, executing them, verifying outputs, and refining or finalizing answers. It uses a two-stage data synthesis approach: GPT-4 generates in... | {
"decision": "Reject"
} |
8DuJ5FK2fa | 2410.05345v1 | Trained Models Tell Us How to Make Them Robust to Spurious Correlation without Group Annotation | {
"content": "## Abstract\n\nAbstract Classifiers trained with Empirical Risk Minimization (ERM) tend to rely on attributes that have high spurious correlation with the target. This can degrade the performance on underrepresented (or minority ) groups that lack these attributes, posing significant challenges for both... | [
{
"id": "IV6HziJSmO",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The paper EVaLS, a method to improve model robustness against spurious correlations without requiring group annotations. EVaLS balances high- and low-loss samples... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"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": "2026-01-14T22:16:00.993994"
} | {
"id": "Gv5N8bDPOP",
"metareview": "The paper having been borderline during the review process, I have read it and made my own idea of its content.\n\nFirst, the paper makes the repeated claim, also repeated here in rebuttals, that its approach is near-optimal. But there is absolutely *no* formal result in the pap... | {
"decision": "Reject"
} |
8EM1A6qfX5 | 2401.14624v3 | Unearthing Large Scale Domain-Specific Knowledge from Public Corpora | {
"content": "## Abstract\n\nAbstract Large language models have demonstrated remarkable potential in various tasks, however, there remains a significant scarcity of open-source models and data for specific domains. Previous works have primarily focused on manually specifying resources and collecting high-quality dat... | [
{
"id": "LtkvKkCioA",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper investigates the problem of bootstrapping domain knowledge from general public corpora in order to reduce cost and time for manual data collection of d... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "1;2;3;2",
"contribution_avg": 2,
"presentation": "1;1;3;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.995063"
} | {
"id": "LUNGGjeVAJ",
"metareview": "**Summary**\n\nFinding reliable data for LLMs is the challenge this paper focus on. The authors propose Retrieve-from-CC, a data collection pipeline consisting of a a) query generation process based on existing language models and b) a document retrieval process based on the gen... | {
"decision": "Reject"
} |
8EfxjTCg2k | 2408.09632v3 | MoDeGPT: Modular Decomposition for Large Language Model Compression | {
"content": "## Abstract\n\nAbstract Large Language Models (LLMs) have reshaped the landscape of artificial intelligence by demonstrating exceptional performance across various tasks. However, substantial computational requirements make their deployment challenging on devices with limited resources. Recently, compre... | [
{
"id": "rJ2rfXKAAO",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a novel model compression method by applying three different matrix decomposition algorithms to three distinct types of computations within Tr... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "3;3;3;2",
"confidence_avg": 2.75,
"soundness": "2;3;3;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 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.996035"
} | {
"id": "dWPmjITj5W",
"metareview": "This paper proposes a new approach to compressing Transformer-based models. The idea is to use a set of particular forms of low-rank matrix factorization for the weight matrices. The authors’ strategy is fairly sophisticated, as it seeks to associate various component operations... | {
"decision": "Accept (Oral)"
} |
8EtSBX41mt | 2403.06833v2 | Can LLMs Separate Instructions From Data? And What Do We Even Mean By That? | {
"content": "## Abstract\n\nAbstract Instruction-tuned Large Language Models (LLMs) show impressive results in numerous practical applications, but they lack essential safety features that are common in other areas of computer science, particularly an explicit separation of instructions and data . This makes them vu... | [
{
"id": "9dz8P5JE8M",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper studies the problem of whether LLMs can separate instructions from data, which is important to the safety of LLMs. Specifically, this paper first intro... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "4;3;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": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.997010"
} | {
"id": "Oc4IuxS6XL",
"metareview": "The paper studies the instruction-data separation problem in LLMs. It introduces a formal measure for this problem, proposes a new benchmark to evaluate the performance of LLMs on this problem, and suggests mitigation strategies for this problem.\n\n+ The paper is well-written.\... | {
"decision": "Accept (Poster)"
} |
8FxELTdwJR | 2403.09066v3 | Hyperparameters in Continual Learning: A Reality Check | {
"content": "## Abstract\n\nAbstract Continual learning (CL) aims to train a model on a sequence of tasks ( i.e. , a CL scenario) while balancing the trade-off between plasticity (learning new tasks) and stability (retaining prior knowledge). The dominantly adopted conventional evaluation protocol for CL algorithms ... | [
{
"id": "lTT5o6aEqi",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper proposes a more rigorous evaluation protocol for continual learning methods, emphasizing generalization to unseen scenarios. In contrast to the traditio... | {
"rating": "3;3;6",
"rating_avg": 4,
"confidence": "5;5;4",
"confidence_avg": 4.666666666666667,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.997881"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8GFoOB7XB4 | 2405.18176v3 | SEMF: Supervised Expectation-Maximization Framework for Predicting Intervals | {
"content": "## Abstract\n\nAbstract This work introduces the Supervised Expectation-Maximization Framework (SEMF), a versatile and model-agnostic approach for generating prediction intervals in datasets with complete or missing data. SEMF extends the Expectation-Maximization algorithm, traditionally used in unsuper... | [
{
"id": "mAnTIUIZTm",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "Quantification of uncertainty of a given prediction made by the models is critical for a variety of downstream applications. One way to measure the uncertainty of... | {
"rating": "3;3;3;3",
"rating_avg": 3,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "3;2;3;2",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;1;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.998576"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8Gqz2opok1 | 2410.09408v1 | C-Adapter: Adapting Deep Classifiers for Efficient Conformal Prediction Sets | {
"content": "## Abstract\n\nAbstract Conformal prediction, as an emerging uncertainty quantification technique, typically functions as post-hoc processing for the outputs of trained classifiers.\nTo optimize the classifier for maximum predictive efficiency, Conformal Training rectifies the training objective with a ... | [
{
"id": "NRgP1CtIN8",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces C-Adapter, an adapter-based tuning method designed to improve the efficiency of conformal predictors without compromising classification accu... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "3;2;3;4",
"contribution_avg": 3,
"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": "2026-01-14T22:16:00.999293"
} | {
"id": "h7BF9gSykn",
"metareview": "The paper introduces the Conformal Adapter (C-Adapter), a new method designed to improve the efficiency of conformal prediction sets while maintaining classification accuracy. This post-processing layer retains label ranking in output logits and optimises conformal prediction ef... | {
"decision": "Accept (Poster)"
} |
8J2djeuNDN | 2406.10521v3 | MALLM-GAN: Multi-Agent Large Language Model as Generative Adversarial Network for Synthesizing Tabular Data | {
"content": "## Abstract\n\nAbstract In the era of big data, access to abundant data is crucial for driving research forward. However, such data is often inaccessible due to privacy concerns or high costs, particularly in healthcare domain. Generating synthetic (tabular) data can address this, but existing models ty... | [
{
"id": "ABAFgmHWUU",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In this paper, the authors propose a novel model that generates synthetic tabular data via a proposed multi-agent LLMs framework. The proposed method aims to hand... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "1;3;2;3",
"contribution_avg": 2.25,
"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": "2026-01-14T22:16:01.000028"
} | {
"id": "ReQKxSEwOV",
"metareview": "## Summary: \nThe paper introduces a novel framework for generating synthetic tabular data using large language models (LLMs) inspired by Generative Adversarial Networks (GANs). The proposed approach leverages the in-context learning capabilities of LLMs without requiring fine-t... | {
"decision": "Reject"
} |
8K36RkrI7N | 2408.09000v2 | Classifier-Free Guidance is a Predictor-Corrector | {
"content": "## Abstract\n\nAbstract We investigate the theoretical foundations of\nclassifier-free guidance (CFG).\nCFG is the dominant method of conditional sampling for text-to-image diffusion models, yet\nunlike other aspects of diffusion, it remains on shaky theoretical footing. In this paper, we first disprove... | [
{
"id": "H8jqxZfO59",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper attempts to understand classifier-free guidance from a theoretical perspective. A special characteristic of classifier-free guidance is that it introdu... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "1;2;3;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": "2026-01-14T22:16:01.000668"
} | {
"id": "sWirr23U8z",
"metareview": "The manuscript studies classifier-free guidance for conditional generation. From a theoretical point of view, the manuscript aims to clarifies some misconception of the classifier-free guidance model in the literature, and also provides some new perspective from the connection w... | {
"decision": "Reject"
} |
8KQzoD5XAr | 2409.12993v1 | CraftRTL: High-quality Synthetic Data Generation for Verilog Code Models with Correct-by-Construction Non-Textual Representations and Targeted Code Repair | {
"content": "## Abstract\n\nAbstract Despite the significant progress made in code generation with large language models, challenges persist, especially with hardware description languages such as Verilog. This paper first presents an analysis of fine-tuned LLMs on Verilog coding, with synthetic data from prior meth... | [
{
"id": "RUca4uytEj",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "This paper does a thorough evaluation of LLMs for verilog code generation. They first analyze existing model performance on Verilog code generation tasks, identif... | {
"rating": "5;5;8;8",
"rating_avg": 6.5,
"confidence": "4;1;3;4",
"confidence_avg": 3,
"soundness": "3;3;4;4",
"soundness_avg": 3.5,
"contribution": "3;2;3;4",
"contribution_avg": 3,
"presentation": "2;1;4;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.001413"
} | {
"id": "eeEiphMgy7",
"metareview": "The paper introduces methods for training code LLMs on an important hardware description language, doing a deep dive on the specific kinds of errors made on those problems and addressing them through correct-by-construction synthetic data generation. The primary strength is that... | {
"decision": "Accept (Poster)"
} |
8LZ1D1yqeg | 2410.18764v1 | Task Calibration: Calibrating Large Language Models on Inference Tasks | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have exhibited impressive zero-shot performance on inference tasks. However, LLMs may suffer from spurious correlations between input texts and output labels, which limits LLMs’ ability to reason based purely on general language understanding. In othe... | [
{
"id": "k17UO3D6xX",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In this paper, the authors propose a calibration strategy for NLI based tasks. This calibration strategy runs in inference time, requiring no modification of the ... | {
"rating": "1;3;10",
"rating_avg": 4.666666666666667,
"confidence": "5;5;3",
"confidence_avg": 4.333333333333333,
"soundness": "1;3;4",
"soundness_avg": 2.6666666666666665,
"contribution": "1;2;4",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.002054"
} | {
"id": "xezdqzaLNs",
"metareview": "This paper introduces Task Calibration (TC), a method to enhance LLM reasoning by balancing reliance on the premise and hypothesis, addressing spurious correlations, and improving zero-shot and few-shot performance across various tasks. However, the approach's applicability is n... | {
"decision": "Reject"
} |
8Livf4oZxz | 2410.02713v2 | Video Instruction Tuning with Synthetic Data | {
"content": "## Abstract\n\nAbstract The development of video large multimodal models (LMMs) has been hindered by the difficulty of curating large amounts of high-quality raw data from the web. To address this, we consider an alternative approach, creating a high-quality synthetic dataset specifically for video inst... | [
{
"id": "bFNBOekudo",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposed a synthetic dataset LLaVA-Video-178K, which consists of 178510 videos with detailed annotations, open-ended questions and multiple-choice ques... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "4;3;5;4",
"confidence_avg": 4,
"soundness": "2;3;3;2",
"soundness_avg": 2.5,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "1;3;3;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.002663"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8Lqb1dbbfa | 2406.01651v3 | FusionDTI: Fine-grained Binding Discovery with Token-level Fusion for Drug-Target Interaction | {
"content": "## Abstract\n\nAbstract Predicting drug-target interaction (DTI) is critical in the drug discovery process. Despite remarkable advances in recent DTI models through the integration of representations from diverse drug and target encoders, such models often struggle to capture the fine-grained interactio... | [
{
"id": "nAP5FzD3Ot",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents FusionDTI, a new model designed to improve drug-target interaction (DTI) predictions. FusionDTI employs a token-level Fusion module to capture... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "5;4;5;4",
"confidence_avg": 4.5,
"soundness": "2;2;3;2",
"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": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.003389"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8NiTKmEzJV | 2410.02711v2 | NETS: A Non-Equilibrium Transport Sampler | {
"content": "## Abstract\n\nAbstract We propose an algorithm, termed the Non-Equilibrium Transport Sampler (NETS), to sample from unnormalized probability distributions. NETS can be viewed as a variant of annealed importance sampling (AIS) based on Jarzynski’s equality, in which the stochastic differential equation ... | [
{
"id": "VtxgQRwO44",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 2,
"summary": "This work presents an approach to sample from densities specified up to a normalizing constant using controlled annealed Langevin dynamics. The approach is first ... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;4;4;2",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "1;3;3;4",
"contribution_avg": 2.75,
"presentation": "2;2;4;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.004167"
} | {
"id": "B6mEK5PUnB",
"metareview": "This study proposes a novel sampling method inspired by insights from nonequilibrium physics to enable efficient sampling from unnormalized distributions. The method employs stochastic differential equations (SDEs) and introduces auxiliary dimensions to reformulate the process w... | {
"decision": "Reject"
} |
8OcM1pTfHm | 2406.03642v1 | Is Free Self-Alignment Possible? | {
"content": "## Abstract\n\nAbstract Aligning pretrained language models (LMs) is a complex and resource-intensive process, often requiring access to large amounts of ground-truth preference data and substantial compute. Are these costs necessary?\nThat is, it is possible to align using only inherent model knowledge... | [
{
"id": "aZMgfNiwuq",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces AlignEZ, a method for aligning large language models (LMs) with human preferences without using additional training data or fine-tuning. The... | {
"rating": "3;5;5",
"rating_avg": 4.333333333333333,
"confidence": "4;3;2",
"confidence_avg": 3,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;3;2",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.005112"
} | {
"id": "zOvEmLJdp9",
"metareview": "ALIGNEZ is a cost-effective approach for aligning language models through self-generated preferences and representation editing, eliminating the need for additional training. However, reviewers have raised concerns about its limited innovation and a lack of clarity in the method... | {
"decision": "Reject"
} |
8Q0beBHq41 | 2409.12889v2 | Can VLMs Play Action Role-Playing Games? Take Black Myth Wukong as a Study Case | {
"content": "## Abstract\n\nAbstract Recently, large language model (LLM)-based agents have made significant advances across various fields. One of the most popular research areas involves applying these agents to video games. Traditionally, these methods have relied on game APIs to access in-game environmental and ... | [
{
"id": "N8peAYZaQ6",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This work builds an AI agent framework, aiming to deal with Action Role-Playing Games. It successfully demonstrates the huge potential of LMM in decision-making. ... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;3;3;4",
"confidence_avg": 3.5,
"soundness": "3;3;2;2",
"soundness_avg": 2.5,
"contribution": "3;2;2;2",
"contribution_avg": 2.25,
"presentation": "2;3;4;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.005637"
} | {
"id": "kIxbdZISU3",
"metareview": "This paper proposes to benchmark VLM agents on games, a promising direction to test agentic reasoning capabilities in a sandbox environment. The reviewers raised several solvable concerns, but the authors did not engage. Thus, the paper remains below the threshold, with encourag... | {
"decision": "Reject"
} |
8ROIRnKloJ | 2410.04081v1 | $\epsilon$-VAE: Denoising as Visual Decoding | {
"content": "## Abstract\n\nAbstract In generative modeling, tokenization simplifies complex data into compact, structured representations, creating a more efficient, learnable space. For high-dimensional visual data, it reduces redundancy and emphasizes key features for high-quality generation.\nCurrent visual toke... | [
{
"id": "zRWzYpjYqB",
"initial_rating": 8,
"confidence": 5,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This paper proposes an autoencoder trained with diffusion loss, together with LPIPS loss and GAN loss applied on the estimated sample from the diffusion decoder. ... | {
"rating": "3;3;5;5;8;8",
"rating_avg": 5.333333333333333,
"confidence": "4;4;4;4;4;5",
"confidence_avg": 4.166666666666667,
"soundness": "2;2;4;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;1;2;4;4",
"contribution_avg": 2.5,
"presentation": "2;3;3;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.006274"
} | {
"id": "gqyutqfuPh",
"metareview": "The submission presents an approach that integrates diffusion decoders into the VAE framework, but reviewers expressed concerns about its limited novelty, primarily focusing on replacing the VAE decoder without introducing substantial technical innovations. While the authors pro... | {
"decision": "Reject"
} |
8Rov0fjpOL | 2407.02551v2 | Breach By A Thousand Leaks: Unsafe Information Leakage in 'Safe' AI Responses | {
"content": "## Abstract\n\nAbstract Vulnerability of Frontier language models to misuse and jailbreaks has prompted the development of safety measures like filters and alignment training in an effort to ensure safety through robustness to adversarially crafted prompts. We assert that robustness is fundamentally ins... | [
{
"id": "wjIVMFxHoy",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper explores the abilities of language models to defend against attacks in which a user query is decomposed into multiple questions. They describe a threat... | {
"rating": "3;3;3;6;6",
"rating_avg": 4.2,
"confidence": "4;3;4;2;3",
"confidence_avg": 3.2,
"soundness": "1;2;2;3;3",
"soundness_avg": 2.2,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "3;3;4;4;3",
"presentation_avg": 3.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.007002"
} | {
"id": "7WUAHLLz01",
"metareview": "The submission \"Breach By A Thousand Leaks: Unsafe Information Leakage in 'Safe' AI Responses\" observes that a number of harmful queries against nominally safe language models can be executed by decomposing harmful queries into sets of harmless queries that convey the same pie... | {
"decision": "Accept (Poster)"
} |
8VG8tpPZhe | 2411.00769v1 | GameGen-$\mathbb{X}$: Interactive Open-world Game Video Generation | {
"content": "## Abstract\n\nAbstract We introduce GameGen- 𝕏 𝕏 \\mathbb{X} blackboard_X , the first diffusion transformer model specifically designed for both generating and interactively controlling open-world game videos.\nThis model facilitates high-quality, open-domain generation by simulating an extensive arr... | [
{
"id": "HsuQ8QPAqT",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In this paper the authors present a new dataset of modern AAA games for the purpose of world model training, which they call the Open-World Video Game Dataset (OG... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;5;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;1;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.008011"
} | {
"id": "lZ00T7MoxD",
"metareview": "This is a nice paper, introducing an excellent dataset for generating game videos that will likely be of much use, and a diffusion-based method for controllable game video generation. The dataset is perhaps the biggest contribution, but the method, while not super novel, is also... | {
"decision": "Accept (Poster)"
} |
8VtGeyJyx9 | 2410.10254v2 | LoLCATs: On Low-Rank Linearizing of Large Language Models | {
"content": "## Abstract\n\nAbstract Recent works show we can linearize large language models (LLMs)—swapping the quadratic attentions of popular Transformer-based LLMs with subquadratic analogs, such as linear attention—avoiding the expensive pretraining costs.\nHowever, linearizing LLMs\noften significantly degrad... | [
{
"id": "sHj78mJ3Ew",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper presents LoLCATs, a method for converting large language models (LLMs) with quadratic attention complexity into models with linear complexity while mai... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;2;4;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.009030"
} | {
"id": "7O6Ou8tHpS",
"metareview": "## Summary\n\nThe paper introduces LoLCATs, a method proposed for improving the efficiency and scalability of large language models (LLMs) by replacing quadratic softmax attention with subquadratic linear attention. LoLCATs use a two-step process: attention transfer, where linea... | {
"decision": "Accept (Poster)"
} |
8WQ7VTfPTl | 2410.12299v1 | Semantics-Adaptive Activation Intervention for LLMs via Dynamic Steering Vectors | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have achieved remarkable performance across many tasks, yet aligning them with desired behaviors remains challenging. Activation intervention has emerged as an effective and economical method to modify the behavior of LLMs. Despite considerable intere... | [
{
"id": "C6jMIheJGR",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a novel Semantics-Adaptive Dynamic Intervention (SADI) technique for Large Language Models (LLMs). Unlike conventional methods using fixed ste... | {
"rating": "3;6;6;6;8",
"rating_avg": 5.8,
"confidence": "5;3;3;3;4",
"confidence_avg": 3.6,
"soundness": "2;3;2;3;4",
"soundness_avg": 2.8,
"contribution": "2;3;3;3;4",
"contribution_avg": 3,
"presentation": "2;3;2;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.009844"
} | {
"id": "0JP5BsOVrj",
"metareview": "This paper is dedicated to aligning large language models (LLMs) with desired behaviors. Most current activation intervention methods rely on fixed steering vectors that lack adaptability to input semantics. To address this, they propose Semantics-Adaptive Dynamic Intervention (... | {
"decision": "Accept (Poster)"
} |
8X3OWi2weV | 2407.11477v1 | XTraffic: A Dataset Where Traffic Meets Incidents with Explainability and More | {
"content": "## Abstract\n\nAbstract Long-separated research has been conducted on two highly correlated tracks: traffic and incidents. Traffic track witnesses complicating deep learning models, e.g., to push the prediction a few percent more accurate, and the incident track only studies the incidents alone, e.g., t... | [
{
"id": "B5lecQIfs6",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces a new dataset named XTraffic, which provides time-series indexes on Traffic data (traffic flow, lane occupancy, and average vehicle speed), ... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "5;3;5;3",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;2;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.011325"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8ZLzw5pIrc | 2410.12214v2 | Order-aware Interactive Segmentation | {
"content": "## Abstract\n\nAbstract Interactive segmentation aims to accurately segment target objects with minimal user interactions.\nHowever, current methods often fail to accurately separate target objects from the background, due to a limited understanding of order , the relative depth between objects in a sce... | [
{
"id": "BNOjfyk5SH",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents Order-Aware Interactive Segmentation (OIS), combining order-aware and object-aware attention to improve segmentation accuracy and efficiency. O... | {
"rating": "5;5;5;5;5;8",
"rating_avg": 5.5,
"confidence": "5;3;5;5;4;4",
"confidence_avg": 4.333333333333333,
"soundness": "3;3;4;2;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;2;2;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.012237"
} | {
"id": "snQmLI1OmK",
"metareview": "The paper receives 4 positive and 2 negative ratings after rebuttal, with 3 upgraded scores. Initially, the reviewers had several concerns about technical motivation/contribution, ablation study, handling multiple objects, robustness to depth maps, and experimental fairness. In ... | {
"decision": "Accept (Poster)"
} |
8ZPLn3GCDb | 2410.02744v1 | Neutral residues: revisiting adapters for model extension | {
"content": "## Abstract\n\nAbstract We address the problem of extending a pretrained large language model to a new domain that was not seen at training time, like adding a language for which the original model has seen no or little training data.\nPopular solutions like fine-tuning or low-rank adaptation are succes... | [
{
"id": "5wNVHgJRRB",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes neutral residues, an improvement on adapters that allows for domain adaptation while preserving the model performance in the original domain. ... | {
"rating": "3;5;5",
"rating_avg": 4.333333333333333,
"confidence": "4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;1;3",
"soundness_avg": 2,
"contribution": "2;3;2",
"contribution_avg": 2.3333333333333335,
"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": "2026-01-14T22:16:01.012927"
} | {
"id": "7s5PiZyOra",
"metareview": "This paper proposes an improved method of domain adaptation through model extension, which preserves the performance of the model on its original dataset. \n\nThe paper has extensive experiments on ideal hyperparameters and data composition, and also does show an improvement in ... | {
"decision": "Reject"
} |
8dzKkeWUUb | 2408.15545v3 | SciLitLLM: How to Adapt LLMs for Scientific Literature Understanding | {
"content": "## Abstract\n\nAbstract Scientific literature understanding is crucial for extracting targeted information and garnering insights, thereby significantly advancing scientific discovery.\nDespite the remarkable success of Large Language Models (LLMs), they face challenges in scientific literature understa... | [
{
"id": "5EIeOPWDKB",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a method to improve scientific instruction following through post-training Qwen models. The main contribution is to collect science textbook... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.014104"
} | {
"id": "cxLKzU2ulu",
"metareview": "This paper presents SciLitLLM, a language model tailored for scientific literature understanding, utilizing a hybrid approach that combines continual pre-training (CPT) with supervised fine-tuning (SFT). The proposed pipeline emphasizes constructing high-quality CPT datasets and... | {
"decision": "Accept (Poster)"
} |
8eNLKk5by4 | 2410.02275v1 | Optimal Strong Regret and Violation in Constrained MDPs via Policy Optimization | {
"content": "## Abstract\n\nAbstract We study online learning in constrained MDPs (CMDPs), focusing on the goal of attaining sublinear strong regret and strong cumulative constraint violation.\nDifferently from their standard (weak) counterparts, these metrics do not allow negative terms to compensate positive ones,... | [
{
"id": "HXVlSfmUEp",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper studies efficient online policy optimization in \"*loop-free*\" constrained MDPs (CMDPs) that slightly generalizes finite-horizon episodic CMDPs, where... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "4;2;3",
"confidence_avg": 3,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.014946"
} | {
"id": "SB2cCk8TlF",
"metareview": "This paper proposes a novel algorithm in constrained MDPs that achieves strong regret and strong violation. The new algorithm is based on policy optimization and uses a primal-dual approach. The new result in this paper resolves an open question raised by prior work on this topi... | {
"decision": "Accept (Poster)"
} |
8eenzfwKqU | 2410.05259v1 | GS-VTON: Controllable 3D Virtual Try-on with Gaussian Splatting | {
"content": "## Abstract\n\nAbstract Diffusion-based 2D virtual try-on (VTON) techniques have recently demonstrated strong performance, while the development of 3D VTON has largely lagged behind.\nDespite recent advances in text-guided 3D scene editing, integrating 2D VTON into these pipelines to achieve vivid 3D VT... | [
{
"id": "1Zo4HgokyO",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a novel approach for achieving 3D virtual try-on (VTON) that addresses current limitations in consistency and spatial relationships when ext... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;5;2",
"confidence_avg": 3.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;1;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.015823"
} | {
"id": "q0Edc6xcbs",
"metareview": "This submission proposes a 3D virtual try-on method that leverages 3D Gaussian Splatting (3DGS) and diffusion model adaptation to address limitations in consistency and spatial relationships, present in existing 2D methods. The approach enables multi-view image editing towards i... | {
"decision": "Reject"
} |
8efAVon0eD | 2410.02735v1 | OOD-Chameleon: Is Algorithm Selection for OOD Generalization Learnable? | {
"content": "## Abstract\n\nAbstract Out-of-distribution (OOD) generalization is challenging because distribution shifts come in many forms.\nA multitude of learning algorithms exist and each can improve performance in specific OOD situations.\nWe posit that much of the challenge of OOD generalization lies in choosi... | [
{
"id": "2XRQeWHrbn",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 2,
"presentation": 1,
"summary": "Authors formulate a new problem setting of predicting the performances of multiple algorithms on a given dataset without training models. They propose a framework... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "3;3;4;3",
"confidence_avg": 3.25,
"soundness": "2;2;1;3",
"soundness_avg": 2,
"contribution": "2;1;2;3",
"contribution_avg": 2,
"presentation": "2;3;1;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.016552"
} | {
"id": "aRRPiDgUVy",
"metareview": "The overall quality is not good enough to make it an ICLR paper.\n\nThe authors should not complain all the three negative reviewers (rating 3) and say the only positive reviewer (rating 6) is the only one providing reasonable comments and suggestions, which is too aggressive an... | {
"decision": "Reject"
} |
8enWnd6Gp3 | 2405.20283v3 | TetSphere Splatting: Representing High-Quality Geometry with Lagrangian Volumetric Meshes | {
"content": "## Abstract\n\nAbstract We introduce TetSphere Splatting, a Lagrangian geometry representation\ndesigned for high-quality 3D shape modeling.\nTetSphere splatting leverages an underused yet powerful geometric primitive – volumetric tetrahedral meshes.\nIt represents 3D shapes by deforming a collection of... | [
{
"id": "BsXPIobOit",
"initial_rating": 3,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a method for reconstructing 3D geometry from multiview images. The geometry is represented by Tetrahedral meshes of sphere topology that are o... | {
"rating": "3;6;6;8;8",
"rating_avg": 6.2,
"confidence": "5;4;2;4;2",
"confidence_avg": 3.4,
"soundness": "3;3;2;4;3",
"soundness_avg": 3,
"contribution": "2;3;3;3;3",
"contribution_avg": 2.8,
"presentation": "3;3;3;3;4",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.017283"
} | {
"id": "uQ1jm3KHY0",
"metareview": "The submission received positive reviews from all the reviewers. The reviewers generally appreciate the clarity, recognize the novelty of the method, and are convinced by the positive experimental results. After reading the paper, the reviewers' comments and the authors' rebutta... | {
"decision": "Accept (Oral)"
} |
8g4XgC8HPF | 2410.13111v1 | Controllable Generation via Locally Constrained Resampling | {
"content": "## Abstract\n\nAbstract Autoregressive models have demonstrated an unprecedented ability at modeling the intricacies\nof natural language.\nHowever, they continue to struggle with generating complex outputs that adhere to\nlogical constraints.\nSampling from a fully-independent distribution subject to a... | [
{
"id": "QXeVfLxJ7D",
"initial_rating": 3,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper studies the problem of sampling from the distribution given by a pretrained language model subject to logical constraints (which could be expressed as ... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "3;5;4;3",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "4;3;2;3",
"contribution_avg": 3,
"presentation": "2;3;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.018256"
} | {
"id": "7dimE3JMQ6",
"metareview": "This paper focuses on sampling from a language model subject to a constraint, i.e. from a distribution proportional to LM(x) * constraint(x). The authors propose using importance sampling where the proposal distribution is constructed via “knowledge compilation” which makes the ... | {
"decision": "Accept (Poster)"
} |
8iH8YHrGTh | 2310.19470v2 | Bridging Lottery Ticket and Grokking: Understanding Grokking from Inner Structure of Networks | {
"content": "## Abstract\n\nAbstract Grokking is the intriguing phenomenon of delayed generalization: initially, a network achieves a memorization solution with perfect training accuracy and limited generalization solution; however, through further training, it eventually attains a generalization solution.\nThis pap... | [
{
"id": "y2b6fNgM9b",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper examines the phenomenon of \"grokking,\" in which neural networks first achieve high training accuracy but poor generalization, then later switch to a ... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "3;4;4;3",
"confidence_avg": 3.5,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;3;3;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.019616"
} | {
"id": "m2rQENhhQk",
"metareview": "This paper studies the connection between lottery tickets and grokking. The authors make the observation that using lottery tickets chosen at the time of test generalization can mitigate grokking challenges. Here, grokking refers to the phenomena that, for certain tasks such as ... | {
"decision": "Reject"
} |
8jOqCcLzeO | 2407.14207v5 | Longhorn: State Space Models are Amortized Online Learners | {
"content": "## Abstract\n\nAbstract Modern large language models are built on sequence modeling via next-token prediction.\nWhile the Transformer remains the dominant architecture for sequence modeling, its quadratic decoding complexity in sequence length poses a major limitation. State-space models (SSMs) present ... | [
{
"id": "TUDspi943d",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "Longhorn formulates state-space models by solving an online regression problem. By designing various online learning objectives, it can induce different linear re... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;3;4;5",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.020363"
} | {
"id": "7GHQ72v8LO",
"metareview": "This paper offers a novel online-learning perspective on SSM design, introducing a novel architecture called Longhorn and demonstrating improved performance over strong baselines like Mamba. All reviewers appreciated the clarity of exposition, the elegance of the theoretical fra... | {
"decision": "Accept (Poster)"
} |
8jvVNPHtVJ | 2410.10166v1 | Automated Filtering of Human Feedback Data for Aligning Text-to-Image Diffusion Models | {
"content": "## Abstract\n\nAbstract Fine-tuning text-to-image diffusion models with human feedback is an effective method for aligning model behavior with human intentions.\nHowever, this alignment process often suffers from slow convergence due to the large size and noise present in human feedback datasets.\nIn th... | [
{
"id": "MZ9IKk4JTd",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces FiFA, an automated data filtering algorithm designed to optimize fine-tuning of text-to-image diffusion models, aligning model behavior more... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "5;4;3;3",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.021147"
} | {
"id": "vCeCsVIWyK",
"metareview": "This paper proposes a method to automatically select high-quality data for diffusion DPO process, which significantly accelerates the training and reduces the GPU hours. With solid experiments, the paper shows its significance in real-world practice. However, the idea of the pap... | {
"decision": "Accept (Poster)"
} |
8kGonpsiHb | 2410.04407v1 | Lens: Rethinking Multilingual Enhancement for Large Language Models | {
"content": "## Abstract\n\nAbstract Despite the growing global demand for large language models (LLMs) that serve users from diverse linguistic backgrounds, most cutting-edge LLMs remain predominantly English-centric. This creates a performance gap across languages, restricting access to advanced AI services for no... | [
{
"id": "L7Zez3BJSf",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a novel method, Lens, to enhance the multilingual capabilities of LLMs. Lens first explores the subspaces of Language-Specific and Language-Ag... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;3;4;5",
"confidence_avg": 4,
"soundness": "3;2;3;4",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.021963"
} | {
"id": "I1ywLYqKAU",
"metareview": "This work proposes LENS (multiLingual Enhancement method based on the hidden\nrepreseNtations within language Space of LLMs), as a new method to improve the multilingual perfromance of LLMs by modifying their internal representation spaces. Specifically, LENS consists of two ste... | {
"decision": "Reject"
} |
8khcyTc4Di | 2406.07983v1 | Meta-Learning Neural Procedural Biases | {
"content": "## Abstract\n\nAbstract The goal of few-shot learning is to generalize and achieve high performance on new unseen learning tasks, where each task has only a limited number of examples available. Gradient-based meta-learning attempts to address this challenging task by learning how to learn new tasks by ... | [
{
"id": "LKS9iUp6x6",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The authors propose to combine several meta-learning methods into a single one, which they dub “Neural Procedural Biases Meta-Learning” (NPBML). The main idea is ... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "4;2;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "4;3;3;2",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.022667"
} | {
"id": "LMfx1AeMXq",
"metareview": "This paper combines gradient-based meta-learning methods for few-shot learning, learning initialisations, optimisers and loss functions together, with empirical improvements upon existing methods. \n\nReviewers largely agree that it is interesting to combine these methods togeth... | {
"decision": "Reject"
} |
8kk9joQCkc | 2409.20427v2 | Sufficient and Necessary Explanations (and What Lies in Between) | {
"content": "## Abstract\n\nAbstract As complex machine learning models continue to find applications in high-stakes decision making scenarios, it is crucial that we can explain and understand their predictions. Post-hoc explanation methods provide useful insights by identifying important features in an input x 𝑥 x... | [
{
"id": "3OqcsipuIK",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "To address vague notions of feature importance in many XAI methods, the paper introduces formal definitions for sufficiency and necessity for local feature-based ... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "4;4;3;4;4",
"confidence_avg": 3.8,
"soundness": "3;2;4;2;3",
"soundness_avg": 2.8,
"contribution": "1;2;3;3;3",
"contribution_avg": 2.4,
"presentation": "3;3;2;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.023356"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8mE8KNHTjd | 2406.01069v1 | UniQA: Unified Vision-Language Pre-training for Image Quality and Aesthetic Assessment | {
"content": "## Abstract\n\nAbstract Image Quality Assessment (IQA) and Image Aesthetic Assessment (IAA) aim to simulate human subjective perception of image visual quality and aesthetic appeal. Existing methods typically address these tasks independently due to distinct learning objectives. However, they neglect th... | [
{
"id": "5LjjYzvo2S",
"initial_rating": 8,
"confidence": 5,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a unified evaluation model to handle both image quality assessment (IQA) and image aesthetic assessment (IAA) tasks simultaneously. Specifical... | {
"rating": "3;5;5;5;5;5;6;6",
"rating_avg": 5,
"confidence": "5;4;5;4;3;4;3;3",
"confidence_avg": 3.875,
"soundness": "2;3;2;3;2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;2;2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;2;3;2;3;3;3",
"presentation_avg": 2.625
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.024447"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8nLGhdBd9e | 2409.16471v1 | Score-based Neural Ordinary Differential Equations for Computing Mean Field Control Problems | {
"content": "## Abstract\n\nAbstract Classical neural ordinary differential equations (ODEs) are powerful tools for approximating the log-density functions in high-dimensional spaces along trajectories, where neural networks parameterize the velocity fields. This paper proposes a system of neural differential equati... | [
{
"id": "5xSYDkTHwn",
"initial_rating": 1,
"confidence": 4,
"soundness": 1,
"contribution": 1,
"presentation": 2,
"summary": "The paper gives a system of ODE equations to solve transport equations for a density $\\rho(t,x)$ based on the knowledge of the score $s(t,x) = \\nabla \\log \\rh... | {
"rating": "1;5;5;6",
"rating_avg": 4.25,
"confidence": "4;2;2;3",
"confidence_avg": 2.75,
"soundness": "1;3;2;3",
"soundness_avg": 2.25,
"contribution": "1;2;3;3",
"contribution_avg": 2.25,
"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": "2026-01-14T22:16:01.025409"
} | {
"id": "Xxly4pqU11",
"metareview": "This paper derives a system of ODE equations to solve density transport based the score function and associated Hessian. A reformulation of second-order mean field control (MFC) problems using the proposed neural ODEs is explored as an application. A strength of the paper is cl... | {
"decision": "Reject"
} |
8o6LdeVi1K | 2410.06467v1 | WAPITI: A Watermark for Finetuned Open-Source LLMs | {
"content": "## Abstract\n\nAbstract Watermarking of large language models (LLMs) generation embeds an imperceptible statistical pattern within texts, making it algorithmically detectable.\nWatermarking is a promising method for addressing potential harm and biases from LLMs, as it enables traceability, accountabili... | [
{
"id": "e9qjrN4z54",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper introduces WAPITI, a watermarking method for fine-tuned open-source LLMs. It embeds watermarks directly into model parameters, ensuring robustness again... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "4;5;4;4",
"confidence_avg": 4.25,
"soundness": "2;4;2;2",
"soundness_avg": 2.5,
"contribution": "2;1;2;3",
"contribution_avg": 2,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.026317"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8pusxkLEQO | 2410.20502v1 | ARLON: Boosting Diffusion Transformers with Autoregressive Models for Long Video Generation | {
"content": "## Abstract\n\nAbstract Text-to-video (T2V) models have recently undergone rapid and substantial advancements. Nevertheless, due to limitations in data and computational resources, achieving efficient generation of long videos with rich motion dynamics remains a significant challenge.\nTo generate high-... | [
{
"id": "5ZgexS6TU6",
"initial_rating": 5,
"confidence": 2,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The manuscript introduces ARLON, a text-to-video framework that efficiently generates high-quality, dynamic, and temporally consistent long videos. By combining A... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "2;3;3;2",
"confidence_avg": 2.5,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.027230"
} | {
"id": "pEezDYcxzo",
"metareview": "This paper explores the generation of long videos guided by an autoregressive language model (LM) to produce conditions for a diffusion transformer (DiT) model. It introduces new techniques to bridge the gap between the LM and the DiT, including a new VQ-VAE that quantizes the D... | {
"decision": "Accept (Poster)"
} |
8q3WIvJhkl | 2405.17403v2 | A Closer Look at Time Steps is Worthy of Triple Speed-Up for Diffusion Model Training | {
"content": "## Abstract\n\nAbstract Training diffusion models is always a computation-intensive task. In this paper, we introduce a novel speed-up method for diffusion model training, called SpeeD , which is based on a closer look at time steps.\nOur key findings are:\ni) Time steps can be empirically divided into ... | [
{
"id": "1FudbLDmDa",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper introduces SpeeD, a novel approach aimed at improving training efficiency for diffusion models. By analyzing the process increment between adjacent tim... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;4;3;5",
"confidence_avg": 4,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "2;2;2;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.028313"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8q9NOMzRDg | 2410.09575v1 | Reconstructive Visual Instruction Tuning | {
"content": "## Abstract\n\nAbstract This paper introduces r ec o nstructive vi s ual in s truction tuning ( Ross ), a family of Large Multimodal Models (LMMs) that exploit vision-centric supervision signals .\nIn contrast to conventional visual instruction tuning approaches that exclusively supervise text outputs, ... | [
{
"id": "trVbgnTC6k",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes a new LMM training approach with an additional branch for input image reconstruction. The results show the model enhances fine-grained comprehe... | {
"rating": "5;5;6;6;6",
"rating_avg": 5.6,
"confidence": "4;4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;4;3;3",
"soundness_avg": 3,
"contribution": "2;2;4;3;3",
"contribution_avg": 2.8,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.029251"
} | {
"id": "IPYmMswsjZ",
"metareview": "This paper introduces a new approach to visual instruction tuning by incorporating an additional reconstruction objective. The proposed method is interesting and demonstrates clear improvements. Reviewers provided overall positive feedback, and the authors submitted strong rebut... | {
"decision": "Accept (Poster)"
} |
8rbkePAapb | 2410.02246v1 | PFGuard: A Generative Framework with Privacy and Fairness Safeguards | {
"content": "## Abstract\n\nAbstract Generative models must ensure both privacy and fairness for Trustworthy AI. While these goals have been pursued separately, recent studies propose to combine existing privacy and fairness techniques to achieve both goals. However, naïvely combining these techniques can be insuffi... | [
{
"id": "2Ca8ez1JL0",
"initial_rating": 3,
"confidence": 3,
"soundness": 1,
"contribution": 1,
"presentation": 2,
"summary": "The paper presents PFGuard, a framework for jointly private and fair generative models. The challenges of naively integrating an unfairness mitigation scheme with... | {
"rating": "3;3;3;5;6",
"rating_avg": 4,
"confidence": "3;4;3;4;3",
"confidence_avg": 3.4,
"soundness": "2;3;1;3;4",
"soundness_avg": 2.6,
"contribution": "1;2;1;2;4",
"contribution_avg": 2,
"presentation": "3;3;2;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.030082"
} | {
"id": "BiWRaRQ8Y6",
"metareview": "The submitted paper introduces \"PFGuard,\" a framework for (image) generation model designed to navigate fairness-privacy-utility trade-offs. Broadly speaking, PFGuard is based on an ensemble of teacher models for balancing fairness and privacy during model training, blending D... | {
"decision": "Accept (Poster)"
} |
8roRgrjbjv | 2410.06716v1 | Guaranteed Generation from Large Language Models | {
"content": "## Abstract\n\nAbstract As large language models (LLMs) are increasingly used across various applications, there is a growing need to control text generation to satisfy specific constraints or requirements. This raises a crucial question: Is it possible to guarantee strict constraint satisfaction in gen... | [
{
"id": "k0Xr2PDqY4",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 1,
"presentation": 3,
"summary": "This paper motivates and studies the problem of constraining LLM generation on logical constraints with guarantees. The authors show that it is often intractable ... | {
"rating": "3;5;8;8",
"rating_avg": 6,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "1;2;4;3",
"contribution_avg": 2.5,
"presentation": "3;2;4;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.031045"
} | {
"id": "bW39JtY1Lm",
"metareview": "## Summary \n\nThis paper addresses the challenge of ensuring strict constraint satisfaction in text generation by large language models while maintaining closeness to the original model’s distribution. The authors define an ideal distribution that satisfies the constraints and ... | {
"decision": "Accept (Poster)"
} |
8sKXFvSCqA | 2410.04703v1 | Neural Fourier Modelling: A Highly Compact Approach to Time-Series Analysis | {
"content": "## Abstract\n\nAbstract Neural time-series analysis has traditionally focused on modeling data in the time domain, often with some approaches incorporating equivalent Fourier domain representations as auxiliary spectral features. In this work, we shift the main focus to frequency representations, modeli... | [
{
"id": "fabQPdEBws",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "- The work concerns time-series analysis, such as forecasting, classification, and anomaly detection.\n- The proposed method neatly decouples data size and repres... | {
"rating": "5;6;6;6;8",
"rating_avg": 6.2,
"confidence": "4;4;3;4;4",
"confidence_avg": 3.8,
"soundness": "3;3;3;3;4",
"soundness_avg": 3.2,
"contribution": "3;3;3;3;3",
"contribution_avg": 3,
"presentation": "2;3;3;4;4",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Desk Rejected Submission",
"venueid": "ICLR.cc/2025/Conference/Desk_Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.031991"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8shi3NhgJp | 2305.14782v3 | IBCL: Zero-shot Model Generation under Stability-Plasticity Trade-offs | {
"content": "## Abstract\n\nAbstract Algorithms that balance the stability-plasticity trade-off are well-studied in the continual learning literature. However, only a few of them focus on obtaining models for specified trade-off preferences. When solving the problem of continual learning under specific trade-offs (C... | [
{
"id": "q780aNtbNj",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 1,
"summary": "The paper, introduces Imprecise Bayesian Continual Learning (IBCL) to address the Continual Learning under Specific Trade-offs (CLuST) problem. Traditional method... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "3;4;3;2",
"confidence_avg": 3,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;1;4;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.032779"
} | {
"id": "zd8If2DOxQ",
"metareview": "The authors propose Imprecise Bayesian Continual Learning (IBCL) for zero-shot model generation under specified stability-plasticity trade-offs. Despite its innovative framing, there are several key weaknesses of this work. The writing is insufficiently rigorous, with unclear de... | {
"decision": "Reject"
} |
8uXkyWFVum | 2408.06663v2 | Amuro and Char: Analyzing the Relationship between Pre-Training and Fine-Tuning of Large Language Models | {
"content": "## Abstract\n\nAbstract The development of large language models leads to the formation of a pre-train-then-align paradigm, in which the model is typically pre-trained on a large text corpus and undergoes a tuning stage to align the model with human preference or downstream tasks.\nIn this work, we inve... | [
{
"id": "PwRkx08hcp",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper investigates the relationship between pre-training and fine-tuning in large language models by fine-tuning multiple intermediate pre-trained model chec... | {
"rating": "3;3;3;5;5",
"rating_avg": 3.8,
"confidence": "5;4;4;3;3",
"confidence_avg": 3.8,
"soundness": "2;2;2;2;3",
"soundness_avg": 2.2,
"contribution": "2;2;2;2;2",
"contribution_avg": 2,
"presentation": "2;3;3;2;2",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.033797"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
8vzMLo8LDN | 2405.02154v4 | Neural Context Flows for Meta-Learning of Dynamical Systems | {
"content": "## Abstract\n\nAbstract Neural Ordinary Differential Equations (NODEs) often struggle to adapt to new dynamic behaviours caused by parameter changes in the underlying system, even when these dynamics are similar to previously observed behaviours. This problem becomes more challenging when the changing p... | [
{
"id": "V86ViDi30b",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "The paper introduces Neural Context Flows, a method for meta learning. The main contributions of the work are focusing on how to combine context vectors in a way ... | {
"rating": "1;5;6;8",
"rating_avg": 5,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "2;2;3;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.034746"
} | {
"id": "n2F9yOtkgL",
"metareview": "This paper proposes a novel meta-learning strategy for modeling dynamical systems across varying environment parameters. The core idea is to introduce environment-specific latent context vectors and expand the vector field using a Taylor series about these context vectors. This ... | {
"decision": "Accept (Poster)"
} |
8w22WLy2R8 | 2409.20163v1 | MemSim: A Bayesian Simulator for Evaluating Memory of LLM-based Personal Assistants | {
"content": "## Abstract\n\nAbstract LLM-based agents have been widely applied as personal assistants, capable of memorizing information from user messages and responding to personal queries.\nHowever, there still lacks an objective and automatic evaluation on their memory capability, largely due to the challenges i... | [
{
"id": "AA1isplcbQ",
"initial_rating": 3,
"confidence": 2,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "- The paper proposed MemSim, a simulator for automatically constructing diverse and rational QA pairs based on the Bayesian Relation Network. \n- The paper also c... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;2;3;3",
"confidence_avg": 3,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "2;3;2;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": "2026-01-14T22:16:01.035561"
} | {
"id": "o1qIMbn5I1",
"metareview": "The authors present a framework to generate new questions from user messages to evaluate model memory capability.\nThe work seems technically sound, with questions about data generation and baselines addressed during rebuttal. The overall novelty and impact of the work however w... | {
"decision": "Reject"
} |
8w8d8j2FCy | 2410.14972v1 | MENTOR: Mixture-of-Experts Network with Task-Oriented Perturbation for Visual Reinforcement Learning | {
"content": "## Abstract\n\nAbstract Visual deep reinforcement learning (RL) enables robots to acquire skills from visual input for unstructured tasks. However, current algorithms suffer from low sample efficiency, limiting their practical applicability. In this work, we present MENTOR, a method that improves both t... | [
{
"id": "nmeUDHcpye",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The authors introduce MENTOR, a visual deep RL algorithm designed to improve sample efficiency in robotic tasks. MENTOR enhances RL agents by replacing traditiona... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"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": "2026-01-14T22:16:01.036277"
} | {
"id": "Nq1ytmGFBq",
"metareview": "This paper introduces MENTOR, a method aimed at addressing sample efficiency and gradient conflict in Visual RL. It achieves this by employing a mixture-of-experts (MoE) network and a task-oriented perturbation mechanism. The paper demonstrates the effectiveness of this approach... | {
"decision": "Reject"
} |
8wIgDG87jn | 2410.15048v1 | MorphAgent: Empowering Agents through Self-Evolving Profiles and Decentralized Collaboration | {
"content": "## Abstract\n\nAbstract Large Language Model (LLM) based multi-agent systems (MAS) have shown promise in tackling complex tasks, but often rely on predefined roles and centralized coordination, limiting their adaptability to evolving challenges. This paper introduces MorphAgent ,\na novel framework for ... | [
{
"id": "cQro3PPjXw",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces MORPHAGENT, a novel framework for decentralized multi-agent collaboration that enhances problem-solving capabilities in complex tasks through... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;4;4;2",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "1;2;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.036983"
} | {
"id": "m4gX8Wqd1D",
"metareview": "The paper introduces MORPHAGENT, a framework for decentralized multi-agent collaboration that enhances problem-solving capabilities in complex tasks through self-evolving profiles and decentralized collaboration. However, the reviewers also pointed out a lack of novelty, insuffi... | {
"decision": "Reject"
} |
8wjWm5jr1w | 2407.10068v1 | Multi-Granularity Semantic Revision for Large Language Model Distillation | {
"content": "## Abstract\n\nAbstract Knowledge distillation plays a key role in compressing the Large Language Models (LLMs), which boosts a small-size student model under large teacher models’ guidance. However, existing LLM distillation methods overly rely on student-generated outputs, which may introduce generati... | [
{
"id": "byfecDPWql",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces three new objectives for distilling generative LLMs at the token, span, and sequence levels:\n\n(token) DAC-KL: it learns additional models ... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "2;4;2;3",
"soundness_avg": 2.75,
"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": "2026-01-14T22:16:01.037599"
} | {
"id": "oiomBR7gmA",
"metareview": "This paper presents a method to Knowledge Distillation from a larger teacher model, enhancing the off-policy method (DistiLLM) through sequence-level correction and regeneration. It introduces two loss functions: Token-level DAC-KL and Span-level Correlation Consistency. The Tok... | {
"decision": "Reject"
} |
8xStV6KJEr | 2411.00359v1 | Constrained Diffusion Implicit Models | {
"content": "## Abstract\n\nAbstract This paper describes an efficient algorithm for solving noisy linear inverse problems using pretrained diffusion models. Extending the paradigm of denoising diffusion implicit models (DDIM), we propose constrained diffusion implicit models (CDIM) that modify the diffusion updates... | [
{
"id": "eokNrFIAnj",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper suggests a new model, Conditional Diffusion Implicit Models(CDIM) for solving linear inverse problem with pretrained diffusion models.\nCDIM can addres... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;2;2;2",
"soundness_avg": 2.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"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": "2026-01-14T22:16:01.038178"
} | {
"id": "XJjg9D6y8T",
"metareview": "The main claim of the paper is that the proposed approach can solve linear inverse problems 10-50 times faster than existing works on conditional diffusion models. The achieved speed-ups are encouraging and were appreciated by the reviewers. However, the contribution over DDIM i... | {
"decision": "Reject"
} |
8xpR7IXcE8 | 2409.20237v1 | Classroom-Inspired Multi-Mentor Distillation with Adaptive Learning Strategies | {
"content": "## Abstract\n\nAbstract We propose ClassroomKD , a novel multi-mentor knowledge distillation framework inspired by classroom environments to enhance knowledge transfer between student and multiple mentors. Unlike traditional methods that rely on fixed mentor-student relationships, our framework dynamica... | [
{
"id": "GIrrnIx93O",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "after reading the comments from other reviewers and corresponding responses, I would like to keep my initial score due to limited novelty and insufficient experim... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "5;4;3;4",
"confidence_avg": 4,
"soundness": "2;2;4;4",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;2;4;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.038828"
} | {
"id": "R7ZA4iOdlj",
"metareview": "This paper introduces a multi-mentor knowledge distillation framework known as ClassroomKD. It includes a Knowledge Filtering Module that ranks and activates high-quality mentors and a Mentoring Module that adjusts each mentor's influence based on the performance gap with the st... | {
"decision": "Reject"
} |
8yEoTBceap | 2410.02477v1 | Learning Diverse Bimanual Dexterous Manipulation Skills from Human Demonstrations | {
"content": "## Abstract\n\nAbstract Bimanual dexterous manipulation is a critical yet underexplored area in robotics. Its high-dimensional action space and inherent task complexity present significant challenges for policy learning, and the limited task diversity in existing benchmarks hinders general-purpose skill... | [
{
"id": "kfRFWnewq9",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces **BiDexHD**, a framework designed to automatically turn a human bimanual manipulation dataset into simulation tasks and learn diverse bimanu... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "3;3;2;2",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;4;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.039487"
} | {
"id": "WAvLeK7EQv",
"metareview": "The submission introduces a framework for learning bimanual dexterous manipulation skills by leveraging human demonstrations. The reviewers acknowledge its novelty in task construction and the teacher-student framework, as well as its promising performance on the TACO dataset. H... | {
"decision": "Reject"
} |
8zJRon6k5v | 2410.05602v1 | Amortized Control of Continuous State Space Feynman-Kac Model for Irregular Time Series | {
"content": "## Abstract\n\nAbstract Many real-world datasets, such as healthcare, climate, and economics, are often collected as irregular time series, which poses challenges for accurate modeling. In this paper, we propose the Amortized Control of continuous State Space Model (ACSSM) for continuous dynamical model... | [
{
"id": "hGZTpu0xu7",
"initial_rating": 8,
"confidence": 2,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "The authors propose ACSSM approach for modeling irregular time series which uses continuous-discrete state space models (CD-SSMs). The authors extend doob's-h tr... | {
"rating": "6;8;8;8",
"rating_avg": 7.5,
"confidence": "2;3;2;2",
"confidence_avg": 2.25,
"soundness": "3;3;4;4",
"soundness_avg": 3.5,
"contribution": "3;3;4;3",
"contribution_avg": 3.25,
"presentation": "2;3;2;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.040427"
} | {
"id": "F5uWM0AuRj",
"metareview": "This paper proposes the Amortized Control of continuous State Space Model for continuous dynamical modeling of time series for irregular and discrete observations. It extends Doob's h-transform to the multi-marginal setting, and defines a variational inference algorithm with a t... | {
"decision": "Accept (Oral)"
} |
8zxGruuzr9 | 2407.12878v3 | Do LLMs have Consistent Values? | {
"content": "## Abstract\n\nAbstract Large Language Models (LLM) technology is constantly improving towards human-like dialogue. Values are a basic driving force underlying human behavior, but little research has been done to study the values exhibited in text generated by LLMs. Here we study this question by turnin... | [
{
"id": "t7axSfjZKO",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 1,
"presentation": 1,
"summary": "This paper examines how different large language models (GPT-4, Gemini Pro, Llama 3.1 8B, Llama 3.1 70B, Gemma 2 9B, and Gemma 2 27B) respond to a 57-item value q... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;2;2;2",
"soundness_avg": 2,
"contribution": "2;2;1;2",
"contribution_avg": 1.75,
"presentation": "2;3;1;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.041396"
} | {
"id": "yFwbD6rJsN",
"metareview": "**Summary:**\n\nThe authors introduce a framework for probing open and closed LLMs' ability to model value rankings and correlations using Schwartz’s Theory of Basic Human Values. They evaluate using 5 types of prompts that either implicitly or explicitly capture value systems (... | {
"decision": "Accept (Poster)"
} |
90DC0IvlSs | 2410.05462v1 | Time, Space and Streaming Efficient Algorithm for Heavy Attentions | {
"content": "## Abstract\n\nAbstract A central problem related to transformers can be stated as follows: given two n × d 𝑛 𝑑 n\\times d italic_n × italic_d matrices Q 𝑄 Q italic_Q and K 𝐾 K italic_K , and a non-negative function f 𝑓 f italic_f , define the matrix A 𝐴 A italic_A as follows: (1) apply the functi... | [
{
"id": "QhZk1uyCEq",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper studies efficient algorithms for attention computation in transformer models. In particular for query $Q$ and key $K$ matrices in $\\mathbb{R}^{n\\time... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "3;4;3;3",
"soundness_avg": 3.25,
"contribution": "3;3;2;2",
"contribution_avg": 2.5,
"presentation": "3;4;3;2",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.042202"
} | {
"id": "ETWSBexiTV",
"metareview": "The paper considers a new algorithm for attention computation under polynomial activation function. There is a need to understand the efficiency of attention mechanisms given their importance in transformer, and the theoretical results here are strong enough that I recommend acc... | {
"decision": "Accept (Poster)"
} |
90Db4RUBc7 | 2410.06846v1 | Joint Fine-tuning and Conversion of Pretrained Speech and Language Models towards Linear Complexity | {
"content": "## Abstract\n\nAbstract Architectures such as Linformer and Mamba have recently emerged as competitive linear time replacements for transformers. However, corresponding large pretrained models are often unavailable, especially in non-text domains. To remedy this, we present a Cross-Architecture Layerwis... | [
{
"id": "VSM99PClel",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper focuses on the issue of the quadratic computational complexity of standard self-attention in transformer models and proposes a way to convert pre-train... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "2;3;4;3",
"contribution_avg": 3,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.042874"
} | {
"id": "riDmRJVhtt",
"metareview": "This paper proposes a approach based on distillation for efficient fine-tuning, with an advantage of linear complexity.\n\nAll reviewers praise the idea interesting, the paper easy to follow, and the experiments solid across multiple modalities and benchmarks.\n\nDespite the rec... | {
"decision": "Accept (Poster)"
} |
90z4EDqcmu | 2410.10745v1 | FlexGen: Flexible Multi-View Generation from Text and Image Inputs | {
"content": "## Abstract\n\nAbstract In this work, we introduce FlexGen, a flexible framework designed to generate controllable and consistent multi-view images, conditioned on a single-view image, or a text prompt, or both. FlexGen tackles the challenges of controllable multi-view synthesis through additional condi... | [
{
"id": "ndf8hceOOw",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper presents FlexGen, a framework designed for multi-view image synthesis using single-view images, text prompts, or both. The core methodology leverages GP... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;5;4;4",
"confidence_avg": 4.25,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.043504"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
9120xQKmcN | 2406.05832v2 | Improving Antibody Design with Force-Guided Sampling in Diffusion Models | {
"content": "## Abstract\n\nAbstract Antibodies, crucial for immune defense, primarily rely on complementarity-determining regions (CDRs) to bind and neutralize antigens, such as viruses. The design of these CDRs determines the antibody’s affinity and specificity towards its target. Generative models, particularly d... | [
{
"id": "ZyhTJ4RSng",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces DiffForce - approach and model allowing to sample from diffusion model with force field guidance. Authors benchmark their method in silico a... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "3;2;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": "2026-01-14T22:16:01.044273"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
94LyPGDi0Y | 2407.14506v2 | On Pre-training of Multimodal Language Models Customized for Chart Understanding | {
"content": "## Abstract\n\nAbstract Recent studies customizing Multimodal Large Language Models (MLLMs) for domain-specific tasks have yielded promising results, especially in the field of scientific chart comprehension.\nThese studies generally utilize visual instruction tuning with specialized datasets to enhance... | [
{
"id": "i0GGVIRCzU",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces a pipeline to create a comprehensive dataset for fine-tuning the proposed MLLM, CHOPINLLM for chart understanding. It highlights that incorpo... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;5;5;4",
"confidence_avg": 4.5,
"soundness": "3;3;2;2",
"soundness_avg": 2.5,
"contribution": "3;3;2;2",
"contribution_avg": 2.5,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.045081"
} | {
"id": "stsEdCgjjg",
"metareview": "The paper introduces CHOPINLLM, a multimodal large language model designed to enhance chart comprehension, particularly for unannotated charts. It presents a data generation pipeline that automatically produces a synthetic dataset tailored for chart understanding tasks and propo... | {
"decision": "Reject"
} |
94kQgWXojH | 2410.02762v1 | Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations | {
"content": "## Abstract\n\nAbstract We investigate the internal representations of vision-language models (VLMs) to address hallucinations, a persistent challenge despite advances in model size and training. We project VLMs’ internal image representations to their language vocabulary and observe more confident outp... | [
{
"id": "i615ZBkfh0",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper addresses the issue of hallucinations in Vision-Language Models (VLMs) by interpreting and editing their internal representations. The authors apply the... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "5;4;4;3",
"confidence_avg": 4,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.045705"
} | {
"id": "aWseziArzs",
"metareview": "This paper proposes an algorithm to erase spurious knowledge from VLMs. The algorithm, coined ProjectAway, relies on Logit Lens to remove information about objects from image representations. The proposed approach is evaluated on several applications: hallucination detection, ha... | {
"decision": "Accept (Poster)"
} |
96GMFXsbJE | 2403.10348v2 | Denoising Task Difficulty-based Curriculum for Training Diffusion Models | {
"content": "## Abstract\n\nAbstract Diffusion-based generative models have emerged as powerful tools in the realm of generative modeling.\nDespite extensive research on denoising across various timesteps and noise levels, a conflict persists regarding the relative difficulties of the denoising tasks.\nWhile various... | [
{
"id": "IL5g252KpU",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper notices that the diffusion training at low noise levels is more challenging than at high ones. \nBased on this observation, the authors propose a curric... | {
"rating": "5;5;6;6;8",
"rating_avg": 6,
"confidence": "3;4;4;4;4",
"confidence_avg": 3.8,
"soundness": "2;3;3;3;3",
"soundness_avg": 2.8,
"contribution": "2;2;3;3;2",
"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": "2026-01-14T22:16:01.046387"
} | {
"id": "GLbWbshdBZ",
"metareview": "This paper provides a neat study of curriculum learning in diffusion models. The idea of training different diffusion models at different training regimes and analyzing the convergence to assess the task difficulty is quite interesting. This is an important contribution as there... | {
"decision": "Accept (Poster)"
} |
96beVMeHh9 | 2206.12525v5 | Causal Identification for Complex Functional Longitudinal Studies | {
"content": "## Abstract\n\nAbstract Real-time monitoring in modern medical research introduces functional longitudinal data, characterized by continuous-time measurements of outcomes, treatments, and confounders. This complexity leads to uncountably infinite treatment-confounder feedbacks, which traditional causal ... | [
{
"id": "n35Ys4D743",
"initial_rating": 3,
"confidence": 2,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "In this paper, authors consider causal inference on time-varying data (functional longitudinal data). They generalize the classical g-computation, inverse probabi... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "3;4;2;3",
"confidence_avg": 3,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "1;1;1;3",
"presentation_avg": 1.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.047131"
} | {
"id": "EPKPD0Z7P5",
"metareview": "The paper provides steps towards causal adjustments with continuous-time data and situations with confounded feedback. While the discrete-time version is well-studied (including work by Robins dating back to the 80s), the continuous-time less so. The paper does provide advances ... | {
"decision": "Accept (Poster)"
} |
96jZFqM5E0 | 2409.09714v1 | Pre-Training for 3D Hand Pose Estimation with Contrastive Learning on Large-Scale Hand Images in the Wild | {
"content": "## Abstract\n\nAbstract We present a contrastive learning framework based on in-the-wild hand images tailored for pre-training 3D hand pose estimators, dubbed HandCLR. Pre-training on large-scale images achieves promising results in various tasks, but prior 3D hand pose pre-training methods have not ful... | [
{
"id": "9HXgKGZeMS",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents a contrastive learning method for the pre-training of 3D hand pose estimation based on large-scale in-the-wild hand images. A parameter-free a... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "5;4;5",
"confidence_avg": 4.666666666666667,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.047894"
} | {
"id": "GU7JW7VBNJ",
"metareview": "This paper proposes a contrastive-learning method for pre-training of 3D hand pose estimation based on large-scale in-the-wild data. \n\nThe three reviewers all appreciate the straightforward nature and effectiveness of the approach, especially as it is demonstrated on large-sca... | {
"decision": "Accept (Poster)"
} |
97dJ3Jp5P4 | 2402.14212v1 | Moonwalk: Inverse-Forward Differentiation | {
"content": "## Abstract\n\nAbstract Backpropagation, while effective for gradient computation, falls short in addressing memory consumption, limiting scalability. This work explores forward-mode gradient computation as an alternative in invertible networks, showing its potential to reduce the memory footprint witho... | [
{
"id": "Cfny1brMLU",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 4,
"summary": "The presented paper proposes a new algorithm for estimating gradients of reversible neural networks, able to decrease the memory footprint compared to standard ba... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;5;3;3",
"confidence_avg": 3.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;1;2;3",
"contribution_avg": 2,
"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": "2026-01-14T22:16:01.048520"
} | {
"id": "n0QlmAilDm",
"metareview": "The paper develops Moonwalk, a forward-mode differentiation method for *submersive* networks, i.e. networks with surjective Jacobians. In particular, all invertible networks are submersive. The method requires computing the gradient with respect to the input first, but then the ... | {
"decision": "Reject"
} |
97rOQDPmk2 | 2410.04870v1 | On the Optimization and Generalization of Two-layer Transformers with Sign Gradient Descent | {
"content": "## Abstract\n\nAbstract The Adam optimizer is widely used for transformer optimization in practice, which makes understanding the underlying optimization mechanisms an important problem.\nHowever, due to the Adam’s complexity, theoretical analysis of how it optimizes transformers remains a challenging t... | [
{
"id": "WBdep4oXzb",
"initial_rating": 8,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper studies the learning dynamics of two-layer transformers optimized with SignGD. It introduces a theoretical framework that categorizes the learning proc... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "3;4;2",
"confidence_avg": 3,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "3;4;2",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.050125"
} | {
"id": "FhQfhndHca",
"metareview": "This work presents a theoretical analysis of the training dynamics of a simplified two-layer transformer using sign gradient descent on a synthetic dataset. They identify four stages in the training dynamics observe that it achieves fast convergence in training loss but generali... | {
"decision": "Accept (Spotlight)"
} |
98ASXp6oPg | 2410.15966v1 | Self-Explained Keywords Empower Large Language Models for Code Generation | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have achieved impressive performance in code generation.\nHowever, due to the long-tail distribution of LLMs’ training data, low-frequency terms are typically underrepresented in the training process. Consequently, LLMs often misunderstand or overlook... | [
{
"id": "qiDgifKlXJ",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a two-stage prompting technique called “Self-Explained Keywords”, that improves the quality of code generated from a variety of LLMs. The te... | {
"rating": "3;3;5",
"rating_avg": 3.6666666666666665,
"confidence": "3;3;4",
"confidence_avg": 3.3333333333333335,
"soundness": "2;2;2",
"soundness_avg": 2,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;2;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.052158"
} | {
"id": "snXyGqkp30",
"metareview": "This paper introduces a novel two-stage prompting method called “Self-Explained Keywords”, aimed at improving code generation quality across a variety of large language models (LLMs). The technique involves: (1) prompting the model to generate descriptions for selected keywords ... | {
"decision": "Reject"
} |
98d7DLMGdt | 2410.03355v1 | LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding | {
"content": "## Abstract\n\nAbstract Auto-Regressive (AR) models have recently gained prominence in image generation, often matching or even surpassing the performance of diffusion models. However, one major limitation of AR models is their sequential nature, which processes tokens one at a time, slowing down genera... | [
{
"id": "zecsjaMhVL",
"initial_rating": 5,
"confidence": 2,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "In this paper, the authors propose LANTERN, a sampling strategy to speed up the image generation without losing too much quality. The method is implemented by acc... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "3;2;4;3",
"confidence_avg": 3,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.052834"
} | {
"id": "F5afnTRddm",
"metareview": "The paper introduces LANTERN, a method that accelerates visual autoregressive models by adapting speculative decoding -- a mechanism originally proposed for large language models -- to the domain of visual autoregressive generation. The proposed approach demonstrates the ability... | {
"decision": "Accept (Poster)"
} |
996aKQIom0 | 2409.06820v1 | PingPong: A Benchmark for Role-Playing Language Models with User Emulation and Multi-Model Evaluation | {
"content": "## Abstract\n\nAbstract We introduce a novel benchmark for evaluating the role-playing capabilities of language models. Our approach leverages language models themselves to emulate users in dynamic, multi-turn conversations and to assess the resulting dialogues. The framework consists of three main comp... | [
{
"id": "uYljdHXaQE",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper introduces PingPong a benchmark that aims to simulate and assess multi-turn interactions using three components Player, Interrogator, and Judge models. ... | {
"rating": "3;3;3;3;5;6",
"rating_avg": 3.8333333333333335,
"confidence": "4;3;4;4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;2;2;3;3",
"soundness_avg": 2.3333333333333335,
"contribution": "1;2;2;2;2;3",
"contribution_avg": 2,
"presentation": "2;3;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": "2026-01-14T22:16:01.053433"
} | {
"id": "PDYbCTuv8d",
"metareview": "This paper introduces a benchmark that includes three types of models: a player model that plays the role of a specific character, an interrogator model that mimics the role of an actual user and interacts with the player model, and a judge model that assesses the quality of the... | {
"decision": "Reject"
} |
9BVMD3keG8 | 2407.01566v1 | A Contextual Online Learning Theory of Brokerage | {
"content": "## Abstract\n\nAbstract We study the role of contextual information in the online learning problem of brokerage between traders.\nAt each round, two traders arrive with secret valuations about an asset they wish to trade.\nThe broker suggests a trading price based on contextual data about the asset.\nTh... | [
{
"id": "5HLRiY6ex2",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper analyzes the problem of online learning of bilateral trading in the contextual setting. The authors provide a comprehensive regret analysis for differe... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;1;2;2",
"contribution_avg": 1.75,
"presentation": "2;3;4;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.054090"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
9BiVepgmWW | 2410.07698v1 | Enhancing Zeroth-order Fine-tuning for Language Models with Low-rank Structures | {
"content": "## Abstract\n\nAbstract Parameter-efficient fine-tuning (PEFT) significantly reduces memory costs when adapting large language models (LLMs) for downstream applications. However, traditional first-order (FO) fine-tuning algorithms incur substantial memory overhead due to the need to store activation val... | [
{
"id": "WcZiIjcysI",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces low-rank zeroth-order optimization algorithms, called LOZO and LOZO-M, for memory-efficient fine-tuning of large language models (LLM). The a... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;4;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.055156"
} | {
"id": "5ezjKOvtFC",
"metareview": "The authors propose a low-rank zero-order (ZO) gradient estimator and introduce a novel algorithm, LOZO, which captures the low-rank gradient structure commonly observed in LLM fine-tuning. The proposed method significantly reduces memory consumption while maintaining performanc... | {
"decision": "Accept (Poster)"
} |
9D2QvO1uWj | 2406.03520v2 | VideoPhy: Evaluating Physical Commonsense for Video Generation | {
"content": "## Abstract\n\nAbstract Recent advances in internet-scale video data pretraining have led to the development of text-to-video generative models that can create high-quality videos across a broad range of visual concepts and styles. Due to their ability to synthesize realistic motions and render complex ... | [
{
"id": "yNXQ61AeDA",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work targets at building a benchmark that can evaluate the physical commonsense for video generation. Multiple physics-related prompts are first generated an... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "2;3;4;4",
"confidence_avg": 3.25,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "2;4;3;4",
"contribution_avg": 3.25,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.055954"
} | {
"id": "mOyFtKEW4v",
"metareview": "The paper addresses an important gap in evaluating physical commonsense in text-to-video (T2V) generation models. The proposed VideoPhy benchmark is insightful, covering diverse physical interactions and revealing significant shortcomings in current models. Reviewers appreciated... | {
"decision": "Accept (Poster)"
} |
9DnKZbOr4r | 2410.18572v1 | Taipan: Efficient and Expressive State Space Language Models with Selective Attention | {
"content": "## Abstract\n\nAbstract Efficient long-context language modeling remains a significant challenge in Natural Language Processing (NLP). While Transformers dominate language tasks, they struggle with long sequences due to quadratic computational complexity in training and linearly scaling memory costs dur... | [
{
"id": "GjQIRE3Lmy",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "The paper presents Taipan, a hybrid model that incorporates attention modules into Mamba-2 (an SSM). Specifically, it proposes to use Selective Attention Layers ... | {
"rating": "3;3;5;5;5",
"rating_avg": 4.2,
"confidence": "5;4;4;4;4",
"confidence_avg": 4.2,
"soundness": "1;2;3;3;3",
"soundness_avg": 2.4,
"contribution": "2;2;2;2;2",
"contribution_avg": 2,
"presentation": "2;1;3;3;3",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.056668"
} | {
"id": "xW6ji5r3KO",
"metareview": "The paper addresses an important challenge in efficient long-context language modeling with a hybrid architecture that combines state-space models and selective attention. While the proposed approach has merit, limited baseline comparisons, and insufficient empirical evidence re... | {
"decision": "Reject"
} |
9EBSEkFSje | 2410.10393v2 | GIFT-Eval: A Benchmark for General Time Series Forecasting Model Evaluation | {
"content": "## Abstract\n\nAbstract Time series foundation models excel in zero-shot forecasting, handling diverse tasks without explicit training. However, the advancement of these models has been hindered by the lack of comprehensive benchmarks. To address this gap, we introduce the G eneral T I me Series F oreca... | [
{
"id": "LqSI8cRZgl",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper targets at a critical and longstanding challenge in time-series forecasting research, lacking a unified, comprehensive, and diverse benchmark for evalu... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;5;5;4",
"confidence_avg": 4.5,
"soundness": "3;2;2;2",
"soundness_avg": 2.25,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"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": "2026-01-14T22:16:01.057369"
} | {
"id": "12731Ki8xx",
"metareview": "The paper introduces GIFT-Eval, a benchmark designed to evaluate time series forecasting models, including statistical, deep learning, and foundation models. It features 28 datasets spanning diverse domains, frequencies, and prediction lengths, alongside a non-leaking pretrainin... | {
"decision": "Reject"
} |
9EfBeXaXf0 | 2409.02135v2 | Optimization by Parallel Quasi-Quantum Annealing with Gradient-Based Sampling | {
"content": "## Abstract\n\nAbstract Learning-based methods have gained attention as general-purpose solvers due to their ability to automatically learn problem-specific heuristics, reducing the need for manually crafted heuristics. However, these methods often face scalability challenges.\nTo address these issues, ... | [
{
"id": "I3MLlO6Dau",
"initial_rating": 8,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In this study, the authors present PQQA, an optimization approach that integrates QQA, gradient-based updates, and parallel run communication. The results indicat... | {
"rating": "3;6;8;8",
"rating_avg": 6.25,
"confidence": "2;4;3;2",
"confidence_avg": 2.75,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "2;3;4;3",
"contribution_avg": 3,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.058402"
} | {
"id": "6HAekDkJxa",
"metareview": "This paper develops a sampling based approach named Parallel Quasi-Quantum Annealing for combinatorial optimization problems. The key ingredients of this approach include a continuous relaxation of the combinatorial optimization problem, an antropic metric to measure discretenes... | {
"decision": "Accept (Poster)"
} |
9EqQC2ct4H | 2407.03153v1 | An Efficient Framework for Crediting Data Contributors of Diffusion Models | {
"content": "## Abstract\n\nAbstract As diffusion models are deployed in real-world settings, data attribution is needed to ensure fair acknowledgment for contributors of high-quality training data and to identify sources of harmful content. Previous work focuses on identifying individual training samples important ... | [
{
"id": "5RGLiQuEOo",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents a framework for attributing the contributions of data providers in diffusion models. The authors propose a framework that efficiently approxim... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "3;3;4",
"confidence_avg": 3.3333333333333335,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.059179"
} | {
"id": "lURY9FOYa6",
"metareview": "This paper looks at data attribution in trained diffusion model through the lens of Shapley values, a common and accepted method originally developed in the economics literature to credit agents’ varying contributions in a cooperative game and has more recently been applied in f... | {
"decision": "Accept (Poster)"
} |
9FRwkPw3Cn | 2406.06560v1 | Inverse Constitutional AI: Compressing Preferences into Principles | {
"content": "## Abstract\n\nAbstract Feedback data plays an important role in fine-tuning and evaluating state-of-the-art AI models. Often pairwise text preferences\nare used: given two texts, human (or AI) annotators select the “better” one. Such feedback data is widely used to align models to human preferences (e.... | [
{
"id": "M8GrCDugi7",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces the Inverse Constitutional AI (ICAI) problem, which seeks to generate a set of principles from a given feedback dataset. These principles se... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;3;3;4",
"contribution_avg": 3,
"presentation": "3;4;2;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.059943"
} | {
"id": "O0kAwlBDQt",
"metareview": "**Summary:**\n\nThis paper highlights the limitations of the current pairwise feedback data in preference optimization. We only know which one is better than another, but don't know \"why\". So, the author introduces a new problem called an ICAI problem, which formulates the int... | {
"decision": "Accept (Poster)"
} |
9FqARW7dwB | 2409.19606v1 | Hyper-Connections | {
"content": "## Abstract\n\nAbstract We present hyper-connections, a simple yet effective method that can serve as an alternative to residual connections. This approach specifically addresses common drawbacks observed in residual connection variants, such as the seesaw effect between gradient vanishing and represent... | [
{
"id": "ve8bWNoGsx",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "The paper introduces Hyper-Connections, a novel extension to residual connections that dynamically adjusts the strength of connections between layers in deep neur... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "5;3;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;3;3;3",
"contribution_avg": 2.5,
"presentation": "1;3;4;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.060810"
} | {
"id": "Rr1GsDFf7b",
"metareview": "The proposed hyper-connections (HCs) are a form of learned architecture that can be optimized to connect different representations across depth and width by summation. In this way they are an extension of residual connections. In addition, HCs can be conditioned on the input to ... | {
"decision": "Accept (Poster)"
} |
9H1uctBWgF | 2409.06277v2 | Ferret: Federated Full-Parameter Tuning at Scale for Large Language Models | {
"content": "## Abstract\n\nAbstract Large Language Models (LLMs) have become indispensable in numerous real-world applications. Unfortunately, fine-tuning these models at scale, especially in federated settings where data privacy and communication efficiency are critical, presents significant challenges. Existing m... | [
{
"id": "lrK1tU4u4f",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "To address the issue of communication overhead in federated learning, the paper proposes using random projection to project local updates into a lower-dimensional... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;4;5",
"confidence_avg": 4.333333333333333,
"soundness": "3;3;2",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.666666666666666... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:01.061667"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
9HK2rHNAhd | 2404.04793v2 | SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget | {
"content": "## Abstract\n\nAbstract Optimizing the Key-Value (KV) cache of the Large Language Model (LLM) has been considered critical to saving the cost of inference. Most of the existing KV-cache compression algorithms attempted to sparsify the sequence of tokens by taking advantage of the different importance of... | [
{
"id": "OMSHRCiw9S",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes SqueezeAttention, a KV-Cache management algorithm that can be combined with KV-Cache eviction policies to further reduce memory footprint and ... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "3;5;4;4",
"confidence_avg": 4,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.062389"
} | {
"id": "QEkgrH3RwP",
"metareview": "The paper presents a practical approach of layer-wise dynamic allocation for KV compression, based on their importance determined by the cosine similarity of embeddings before and after self-attention layers. The method is compatible with other sequence-based compression algori... | {
"decision": "Accept (Poster)"
} |
9Hxdixed7p | 2406.07327v1 | 3D-Properties: Identifying Challenges in DPO and Charting a Path Forward | {
"content": "## Abstract\n\nAbstract Aligning large language models (LLMs) with human preference has recently gained tremendous attention, with the canonical yet costly RLHF-PPO and the simple and straightforward Direct Preference Optimization (DPO) as two examples. Despite the efficiency, DPO has rarely be used in ... | [
{
"id": "ne9wyL1EZ4",
"initial_rating": 6,
"confidence": 3,
"soundness": 4,
"contribution": 2,
"presentation": 3,
"summary": "The paper titled \"3D-Properties: Identifying Challenges in DPO and Charting a Path Forward\" presents a thorough analysis of the DPO method used for aligning LLM... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "4;3;1",
"confidence_avg": 2.6666666666666665,
"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": "2026-01-14T22:16:01.063108"
} | {
"id": "fu8rUmDPOX",
"metareview": "This paper investigates key limitations of Direct Preference Optimization (DPO) in aligning language models, identifying three critical properties termed “3D-properties”: drastic drops in rejected response likelihood, degradation into response suppression, and dispersion effects... | {
"decision": "Accept (Poster)"
} |
9I6UOIfbwf | 2311.11642v3 | Video Face Re-Aging: Toward Temporally Consistent Face Re-Aging | {
"content": "## Abstract\n\nAbstract Video face re-aging deals with altering the apparent age of a person to the target age in videos. This problem is challenging due to the lack of paired video datasets maintaining temporal consistency in identity and age. Most re-aging methods process each image individually witho... | [
{
"id": "OkW5oa9t8h",
"initial_rating": 3,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents a novel approach to video face re-aging, focusing on altering the apparent age of individuals in videos while maintaining temporal consistency.... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "5;5;3;5",
"confidence_avg": 4.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:01.063891"
} | {
"id": "ImaG6nlyGx",
"metareview": "The authors have not been able to convince three reviewers (duBS, 9tWA, azHF) towards the positive side; all these three reviewers agreed this work needs extra efforts to reach the acceptance bar of the ICLR. Thus I am inclined towards not accepting this draft at this stage. Tha... | {
"decision": "Reject"
} |
9JCNPFL1f9 | 2407.13766v2 | Visual Haystacks: A Vision-Centric Needle-In-A-Haystack Benchmark | {
"content": "## Abstract\n\nAbstract Large Multimodal Models (LMMs) have made significant strides in visual question-answering for single images. Recent advancements like long-context LMMs have allowed them to ingest larger, or even multiple, images. However, the ability to process a large number of visual tokens do... | [
{
"id": "Z4oczHY6Rc",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors presents Visual Haystacks (VHs), a new vision centric benchmark designed to assess the performance of Large Multimodal Models (LMMs) in the multi-imag... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "3;4;4",
"confidence_avg": 3.6666666666666665,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;2;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:01.064654"
} | {
"id": "Y23ZeS3Cdj",
"metareview": "The submission introduces a new benchmark \"Visual Haystacks\" for measuring the quality of large multimodal models (LMMs) at the task of multi-image reasoning. Along with the benchmark dataset, it also introduces a RAG framework (MIRAGE) that can process a magnitude larger numb... | {
"decision": "Accept (Poster)"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.