paper_id string | arxiv_id string | title string | markdown dict | reviews list | scores dict | metadata dict | meta_review dict | decision dict |
|---|---|---|---|---|---|---|---|---|
1NkrxqY4jK | 2406.14144v1 | Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) excel in various capabilities but also pose safety risks such as generating harmful content and misinformation, even after safety alignment. In this paper, we explore the inner mechanisms of safety alignment from the perspective of mechanistic interpr... | [
{
"id": "x1qZEX5x9x",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a novel methodology for identifying specific MLP neurons that contribute to safety alignment in large language models. The authors present t... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "3;2;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.66666666666666... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.628531"
} | {
"id": "zXvvdMTr9O",
"metareview": "The recommendation is based on the reviewers' comments, the area chair's evaluation, and the author-reviewer discussion. \n\nWhile the reviewers see some merits in using a mechanistic interpretability approach to study safety neurons in LLMs, this submission should not be accept... | {
"decision": "Reject"
} |
1OyE9IK0kx | 2406.10625v2 | On the Hardness of Faithful Chain-of-Thought Reasoning in Large Language Models | {
"content": "## Abstract\n\nAbstract As Large Language Models (LLMs) are increasingly being employed in real-world applications in critical domains such as healthcare, it is important to ensure that the Chain-of-Thought (CoT) reasoning generated by these models faithfully captures their underlying behavior. While LL... | [
{
"id": "pRqB7bSx02",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In recent years, there have been concerted effort in making language models more faithful and robust with methods such as finetuning, in-context learning and acti... | {
"rating": "3;3;3;5;6;8",
"rating_avg": 4.666666666666667,
"confidence": "3;3;4;4;4;2",
"confidence_avg": 3.3333333333333335,
"soundness": "2;3;3;2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "1;2;2;1;3;3",
"contribution_avg": 2,
"presentation": "3;2;3;2;3;4",
"presentation_avg": 2.8... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.629223"
} | {
"id": "NLlBoedP6i",
"metareview": "The paper examined the Chain-of-Thought (CoT) behavior in large language models (LLMs) through experiments involving in-context learning, fine-tuning, and activation editing. This work combines the three aforementioned techniques to improve LLMs in producing what is referred to ... | {
"decision": "Reject"
} |
1Q2t6D4dK6 | 2408.01541v1 | Guardians of Image Quality: Benchmarking Defenses Against Adversarial Attacks on Image-Quality Metrics | {
"content": "## Abstract\n\nAbstract In the field of Image Quality Assessment (IQA), the adversarial robustness of the metrics poses a critical concern. This paper presents a comprehensive benchmarking study of various defense mechanisms in response to the rise in adversarial attacks on IQA. We systematically evalua... | [
{
"id": "fiDPswmMDx",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper aims to benchmark and evaluate the robustness of 30 different adversarial defense methods against 14 adversarial attacks regarding IQA metrics. It empha... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "5;5;4;3",
"confidence_avg": 4.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;2;3;1",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.629889"
} | {
"id": "1xH0j4ZKCr",
"metareview": "The authors present a comprehensive study benchmarking the robustness of 29 defense methods for 14 adversarial attacks against 9 different image quality assessment (IQA) metrics. While the study is comprehensive and of value to the IQA community, there is too little new insight ... | {
"decision": "Reject"
} |
1Qpt43cqhg | 2405.20445v2 | Fully-inductive Node Classification on Arbitrary Graphs | {
"content": "## Abstract\n\nAbstract Foundation models that can perform inference on any new task without requiring specific training have revolutionized machine learning in vision and language applications. However, applications involving graph-structured data remain a tough nut for foundation models, due to challe... | [
{
"id": "IwyEWtfHt7",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 2,
"summary": "This paper studies the problem of fully inductive node classification, where limited parameters are learned from one small graph, and inference other unseen graph... | {
"rating": "3;6;6;6",
"rating_avg": 5.25,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "2;4;4;4",
"contribution_avg": 3.5,
"presentation": "2;3;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.630552"
} | {
"id": "2gxUVuBD1J",
"metareview": "This paper proposes a general graph neural network that can be applied to new graphs which may have different feature and label space. Authors show that the approach outperforms a wide range of methods. While some reviewers have concerns on the selection of baseline methods and ... | {
"decision": "Accept (Poster)"
} |
1R5BcYS8EC | 2405.19653v2 | SysCaps: Language Interfaces for Simulation Surrogates of Complex Systems | {
"content": "## Abstract\n\nAbstract Surrogate models are used to predict the behavior of complex energy systems that are too expensive to simulate with traditional numerical methods.\nOur work introduces the use of language descriptions, which we call “system captions” or SysCaps, to interface with such surrogates.... | [
{
"id": "soJZYufbl3",
"initial_rating": 5,
"confidence": 2,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper describes a set of lightweight models to model complex energy systems, using an LLM to generate prompts and a encoder and bidirectional time-series mod... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "3;2;2",
"confidence_avg": 2.3333333333333335,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;2",
"contribution_avg": 2,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.631299"
} | {
"id": "iDJFr5vfne",
"metareview": "Reviewers liked that the paper solves the well-motivated problem of interacting with complex energy system (CES) using natural language and the strength of the results. They did not evaluate much of the technical aspects and provided lower confidence reviews: 38tw: could use BER... | {
"decision": "Accept (Poster)"
} |
1S7kpbfgq9 | 2411.04512v1 | Normalized Space Alignment: A Versatile Metric for Representation Analysis | {
"content": "## Abstract\n\nAbstract We introduce a manifold analysis technique for neural network representations. Normalized Space Alignment (NSA) compares pairwise distances between two point clouds derived from the same source and having the same size, while potentially possessing differing dimensionalities. NSA... | [
{
"id": "ci0XfrMIh4",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces Normalized Space Alignment(NSA), a new manifold analysis technique designed to compare neural network representations; NSA compares pairwise ... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;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": "2026-01-14T22:16:00.632237"
} | {
"id": "T6tpu0Sylw",
"metareview": "The authors propose normalized space alignment (NSA) as an analysis technique for neural network representations. NSA combines the global NSA (GNSA) which compares the pairwise Euclidean distances in the two representations, and local NSA (LNSA) which measures dissimilarity from... | {
"decision": "Reject"
} |
1ThYY28HXg | 2411.02319v2 | GenXD: Generating Any 3D and 4D Scenes | {
"content": "## Abstract\n\nAbstract Recent developments in 2D visual generation have been remarkably successful. However, 3D and 4D generation remain challenging in real-world applications due to the lack of large-scale 4D data and effective model design. In this paper, we propose to jointly investigate general 3D ... | [
{
"id": "tX1EIQqCUB",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper proposes GENXD, a latent diffusion model for 3D/4D generation of objects or scenes. Specifically, it adopts masked latent conditions to support various... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "5;5;4;4",
"confidence_avg": 4.5,
"soundness": "1;2;3;3",
"soundness_avg": 2.25,
"contribution": "2;3;4;3",
"contribution_avg": 3,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.633077"
} | {
"id": "Fr3pGPDsXD",
"metareview": "This paper introduces an approach for sequential image generation that can properly depict 3D and 4D scenes. The key idea is to investigate camera and object movements jointly, which leads to curating real-world video to make a new CamVid-30k dataset. The proposed pipeline GenXD... | {
"decision": "Accept (Poster)"
} |
1Uem0nAWK0 | 2410.19206v1 | Inference time LLM alignment in single and multidomain preference spectrum | {
"content": "## Abstract\n\nAbstract Aligning Large Language Models (LLM) to address subjectivity and nuanced preference levels requires adequate flexibility and control, which can be a resource-intensive and time-consuming procedure. Existing\ntraining-time alignment methods require full re-training when a change i... | [
{
"id": "FKVnKsAitC",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces a novel approach for adjusting Large Language Model (LLM) behaviors during inference time using Alignment Vectors (AV). The key innovation i... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "3;2;3;2",
"soundness_avg": 2.5,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"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": "2026-01-14T22:16:00.634020"
} | {
"id": "ZhWuriA4We",
"metareview": "The paper studies the adjustment of LLM behavior at inference time using alignment vectors (AV), defined as the difference between the weights of a model aligned to a particular dimension. The paper creates a synthetic dataset by querying Claude-3-Sonnet to create personas for t... | {
"decision": "Reject"
} |
1X85iw7tqY | 2410.11963v1 | CtrlSynth: Controllable Image Text Synthesis for Data-Efficient Multimodal Learning | {
"content": "## Abstract\n\nAbstract Pretraining robust vision or multimodal foundation models ( e.g . , CLIP) relies on large-scale datasets that may be noisy, potentially misaligned, and have long-tail distributions. Previous works have shown promising results in augmenting datasets by generating synthetic samples... | [
{
"id": "KHxCbnXskn",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors propose a controllable image-text generation pipeline that can augment data to improve CLIPs image retrieval, classification, and compositional perfor... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"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": "2026-01-14T22:16:00.634712"
} | {
"id": "DZSCvY6olX",
"metareview": "The submission proposes CtrlSynth, a controllable pipeline to generate synthetic images for representation learning. The broad workings are as follows: 1) Given a real image and optional associated caption, image tags are generated using a vision model; 2) Using an LLM, a new ca... | {
"decision": "Reject"
} |
1YYp1rPRlm | 2402.05525v2 | Differentially Private Deep Model-Based Reinforcement Learning | {
"content": "## Abstract\n\nAbstract We address private deep offline reinforcement learning (RL), where the goal is to train a policy on standard control tasks that is differentially private (DP) with respect to individual trajectories in the dataset. To achieve this, we introduce PriMORL , a model-based RL algorith... | [
{
"id": "KWaeiL1Uzt",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This work extends model-based offline reinforcement learning, namely MORL, to its differentially private variant, namely PriMORL. The work intends to guarantee tr... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "4;4;2",
"confidence_avg": 3.3333333333333335,
"soundness": "3;2;2",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.635492"
} | {
"id": "vGxhUrGpe0",
"metareview": "This paper studies the problem of differentially private learning in the context of offline RL. The primary concern with the paper is on (1) novelty: extending DP to offline RL is straight-forward conceptually, especially using a model-based method. One can simply apply a DP sup... | {
"decision": "Reject"
} |
1Z3C49JQVf | 2407.10825v2 | Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks | {
"content": "## Abstract\n\nAbstract Deep neural networks are vulnerable to backdoor attacks, a type of adversarial attack that poisons the training data to manipulate the behavior of models trained on such data.\nClean-label attacks are a more stealthy form of backdoor attacks that can perform the attack without ch... | [
{
"id": "L4mmxp0nOO",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes a method for enhancing clean-label backdoor attacks on deep neural networks. Unlike traditional clean-label attacks that apply triggers randoml... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;1;2;3",
"contribution_avg": 2,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.636222"
} | {
"id": "vHk1c9RqcF",
"metareview": "This paper examines clean-label backdoor attacks in a highly constrained setting, where the attacker only has access to training data from the target class and lacks prior knowledge of the victim model, training process, or other classes. In this senario, the authors manage to p... | {
"decision": "Accept (Poster)"
} |
1c73HCZpbo | 2406.14214v6 | REVEAL-IT: REinforcement learning with Visibility of Evolving Agent poLicy for InTerpretability | {
"content": "## Abstract\n\nAbstract Understanding the agent’s learning process, particularly the factors that contribute to its success or failure post-training, is crucial for comprehending the rationale behind the agent’s decision-making process. Prior methods clarify the learning process by creating a structural... | [
{
"id": "Cq1yTvqTcG",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 1,
"summary": "The paper presents an interpretability framework to understand an agent’s learning process in complex tasks (e.g., ALFWorld) through a GNN-based explainer. This m... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "2;4;3;4",
"confidence_avg": 3.25,
"soundness": "2;2;2;2",
"soundness_avg": 2,
"contribution": "2;1;2;2",
"contribution_avg": 1.75,
"presentation": "1;1;2;2",
"presentation_avg": 1.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.637231"
} | {
"id": "6cR3ba6wnU",
"metareview": "The reviewers generally agree that this paper, in its current state, is unfit for publication due to the unclear and, in many parts, incomplete presentation. I encourage the authors to clarify their writing. The authors motivate the method as an interpretability method, but ulti... | {
"decision": "Reject"
} |
1dUdNzLJRF | 2410.03608v1 | TICKing All the Boxes: Generated Checklists Improve LLM Evaluation and Generation | {
"content": "## Abstract\n\nAbstract Given the widespread adoption and usage of Large Language Models (LLMs), it is crucial to have flexible and interpretable evaluations of their instruction-following ability.\nPreference judgments between model outputs have become the de facto evaluation standard, despite distilli... | [
{
"id": "lKdXid6du9",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The authors propose TICK, a method that uses LLMs to decompose instructions into checklists composed of several YES/NO choices to address limitations in standard ... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "3;4;5;4",
"confidence_avg": 4,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"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": "2026-01-14T22:16:00.638091"
} | {
"id": "0XzbUX96XZ",
"metareview": "The authors propose TICK, a method that leverages LLMs to decompose instructions into checklists of yes/no questions. This idea aligns well with assessment practices. The paper conducts comprehensive automated and manual consistency experiments to evaluate and highlight the adva... | {
"decision": "Reject"
} |
1dkVCX4jlH | 2405.11566v1 | Uncertainty-Aware PPG-2-ECG for Enhanced Cardiovascular Diagnosis using Diffusion Models | {
"content": "## Abstract\n\nAbstract Analyzing the cardiovascular system condition via Electrocardiography (ECG) is a common and highly effective approach, and it has been practiced and perfected over many decades.\nECG sensing is non-invasive and relatively easy to acquire, and yet it is still cumbersome for holter... | [
{
"id": "2iSw4ygVcv",
"initial_rating": 5,
"confidence": 5,
"soundness": 4,
"contribution": 3,
"presentation": 2,
"summary": "This paper presents \"Uncertainty-Aware PPG-2-ECG (UA-P2E),\" a novel framework that employs diffusion models to convert photoplethysmography (PPG) signals into e... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "5;3;4",
"confidence_avg": 4,
"soundness": "2;3;4",
"soundness_avg": 3,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "3;3;2",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.638913"
} | {
"id": "3f3nnIJlxa",
"metareview": "This paper presents an approach to generate ECG from PPG for cardiovascular disease detection. However, although the PPG and ECG signals are inter-related, there is a lack of strong evidence that we can use PPG to detect cardio diseases. As pointed out by the authors, the conver... | {
"decision": "Reject"
} |
1eI236MqEA | 2403.11627v2 | LoRA-Composer: Leveraging Low-Rank Adaptation for Multi-Concept Customization in Training-Free Diffusion Models | {
"content": "## Abstract\n\nAbstract Customization generation techniques have significantly advanced the synthesis of specific concepts across varied contexts. Multi-concept customization emerges as the challenging task within this domain. Existing approaches often rely on training a fusion matrix of multiple Low-Ra... | [
{
"id": "99IaYRKAfq",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 1,
"presentation": 2,
"summary": "The paper presents a modified LORA-based multiple-concept generation model. By introducing three loss functions, the phenomenon of concept vanishing and confusion... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "5;4;4;3",
"confidence_avg": 4,
"soundness": "3;1;3;3",
"soundness_avg": 2.5,
"contribution": "2;1;4;3",
"contribution_avg": 2.5,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.639640"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
1eQT9OzfNQ | 2401.03462v3 | Long Context Compression with Activation Beacon | {
"content": "## Abstract\n\nAbstract Long context compression is a critical research problem due to its significance in reducing the high computational and memory costs associated with LLMs. In this paper, we propose Activation Beacon, a plug-in module for transformer-based LLMs that targets effective, efficient, an... | [
{
"id": "5xNgFZh6Re",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The paper introduces “Activation Beacon,” a plug-in module to conduct long-context compression for LLMs. The proposed approach progressively compresses the activa... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;5;3;3",
"confidence_avg": 3.75,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.640540"
} | {
"id": "254OY4I16g",
"metareview": "This paper proposes a context compression method for transformer-based LLMs. The method progressively compresses the key and value activations for all layers into beacon tokens. The paper \bevaluates its benefits for quality and efficiency in long-context tasks. \n\nAll reviewer... | {
"decision": "Accept (Poster)"
} |
1gqR7yEqnP | 1610.02351v4 | Pan for gold | {
"content": "## Abstract\n\nAbstract Many contemporary large-scale applications involve building\ninterpretable models linking a large set of potential covariates to\na response in a nonlinear fashion, such as when the response is\nbinary. Although this modeling problem has been extensively studied,\nit remains uncl... | [
{
"id": "9Y3qqGVN0X",
"initial_rating": 1,
"confidence": 4,
"soundness": 1,
"contribution": 1,
"presentation": 2,
"summary": "This paper explores the finding that training neural networks with random labels leads to substantial performance improvements in comparison to randomly initialis... | {
"rating": "1;1;3;3;3",
"rating_avg": 2.2,
"confidence": "4;4;3;4;4",
"confidence_avg": 3.8,
"soundness": "1;1;2;2;1",
"soundness_avg": 1.4,
"contribution": "1;1;2;2;2",
"contribution_avg": 1.6,
"presentation": "1;2;2;3;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.641677"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
1i6lkavJ94 | 2410.01660v1 | Conformal Generative Modeling with Improved Sample Efficiency through Sequential Greedy Filtering | {
"content": "## Abstract\n\nAbstract Generative models lack rigorous statistical guarantees for their outputs and are therefore unreliable in safety-critical applications. In this work, we propose S equential Co nformal P r e diction for Gen erative Models ( SCOPE-Gen ), a sequential conformal prediction method prod... | [
{
"id": "AW5VZrfOX8",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This fairly well-explicated paper considers the following question of much recent interest: how could we obtain some semblance of guarantees for generative models... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "2;3;1;4",
"confidence_avg": 2.5,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"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": "2026-01-14T22:16:00.642896"
} | {
"id": "v3c2X2vyvE",
"metareview": "This paper propose Sequential Conformal Prediction for Generative Models (SCOPE-Gen), that utilise conformal prediction techniques that operates sequentially, to obtain guarantees and controls for generative models. The proposed approach not only provide practical algorithmic gu... | {
"decision": "Accept (Poster)"
} |
1iuaxjssVp | 2406.11975v1 | Fast Uncovering of Protein Sequence Diversity from Structure | {
"content": "## Abstract\n\nAbstract We present InvMSAFold, a method for generating a diverse set of protein sequences that fold into a single structure. For a given structure, InvMSAFold defines a probability distribution over the space of sequences, capturing the amino acid covariances observed in Multiple Sequenc... | [
{
"id": "kAaxgCBidR",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "InvMSAFold is an inverse folding method that is optimized for diversity and speed. The general idea is to use a neural net to predict from an input structure and ... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "5;4;3;3",
"confidence_avg": 3.75,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;4;4",
"contribution_avg": 3.25,
"presentation": "3;4;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.643627"
} | {
"id": "EaLzjC7cYC",
"metareview": "The paper presents InvMSAFold, an inverse folding method designed to generate diverse protein sequences from a given structure.\n\nThe reviews identified strengths in the speed of the method and the greater diversity that more effectively capture the correlations between residue... | {
"decision": "Accept (Spotlight)"
} |
1lB5ErmIY0 | 2410.14632v2 | Diverging Preferences: When do Annotators Disagree and do Models Know? | {
"content": "## Abstract\n\nAbstract We examine diverging preferences in human-labeled preference datasets. We develop a taxonomy of disagreement sources spanning 10 categories across four high-level classes—task underspecification, response style, refusals, and annotation errors. We find that the majority of disagr... | [
{
"id": "V6A2C6JbRu",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper discusses a proposed distributional reward model aimed at addressing the issues of distinguishing between divided preferences and high-agreement prefere... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;3;5",
"confidence_avg": 4,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "3;2;2;3",
"contribution_avg": 2.5,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.644430"
} | {
"id": "buExeX96rw",
"metareview": "This work examines diverging preferences in human-labeled preference datasets used for training language models. The authors claim that current reward modeling approaches fail to distinguish between high-agreement and diverging preferences, and then propose a taxonomy of disagre... | {
"decision": "Reject"
} |
1mXufFuv95 | 2405.18540v1 | Learning Diverse Attacks on Large Language Models for Robust Red-Teaming and Safety Tuning | {
"content": "## Abstract\n\nAbstract Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automa... | [
{
"id": "sbGn0d7v5Q",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This paper applies GFlowNet to the problem of automated red-teaming, achieving a favorable balance between attack success rate and attack diversity. A model is tr... | {
"rating": "3;5;8;8;8",
"rating_avg": 6.4,
"confidence": "3;4;2;4;4",
"confidence_avg": 3.4,
"soundness": "2;3;3;3;4",
"soundness_avg": 3,
"contribution": "1;2;4;2;3",
"contribution_avg": 2.4,
"presentation": "3;3;4;3;4",
"presentation_avg": 3.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.645143"
} | {
"id": "P8XGZa8mZk",
"metareview": "This paper applies GFlowNet to the problem of automated red-teaming, reaching a favorable balance between attack success rate and attack diversity. The authors are encouraged to include scalable evaluation and thorough results analysis in the final version.",
"additional_comme... | {
"decision": "Accept (Poster)"
} |
1ou5noWgHM | 2310.00646v2 | Source Attribution for Large Language Model-Generated Data | {
"content": "## Abstract\n\nAbstract The impressive performances of large language models (LLMs) and their immense potential for commercialization have given rise to serious concerns over the intellectual property (IP) of their training data. In particular, the synthetic texts generated by LLMs may infringe the IP o... | [
{
"id": "lYeqYq8sSg",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The authors introduce a framework named WASA (Watermarking for Source Attribution) that embeds unique, imperceptible watermarks into the data used for training LL... | {
"rating": "3;3;5;5;6",
"rating_avg": 4.4,
"confidence": "4;4;4;3;4",
"confidence_avg": 3.8,
"soundness": "2;2;2;2;2",
"soundness_avg": 2,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "2;2;3;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.646043"
} | {
"id": "SSLDR53mVq",
"metareview": "The paper received five review scores, three of which were still negative after rebuttal. Although the authors addressed some of the reviewers' concerns during the rebuttal phase (such as the differences from existing methods and missing some baseline methods), the reviewers sti... | {
"decision": "Reject"
} |
1pXzC30ry5 | 2401.10228v1 | RMP-SAM: Towards Real-Time Multi-Purpose Segment Anything | {
"content": "## Abstract\n\nAbstract Advanced by transformer architecture, vision foundation models (VFMs) achieve remarkable progress in performance and generalization ability. Segment Anything Model (SAM) is one remarkable model that can achieve generalized segmentation. However, most VFMs cannot run in real-time,... | [
{
"id": "GDrfMprzci",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper presents a real-time, versatile segmentation model capable of interactive segmentation, panoptic segmentation, and video instance segmentation. \nWhile ... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.646953"
} | {
"id": "9Dp9aNBnmX",
"metareview": "This paper presents RAP-SAM, a real-time, versatile segmentation model capable of handling interactive segmentation, panoptic segmentation, and video instance segmentation. The model retains the SAM encoder-decoder structure while incorporating an efficient encoder and adapter t... | {
"decision": "Accept (Oral)"
} |
1qGkuxI9UX | 2406.00888v1 | Aligning Language Models with Demonstrated Feedback | {
"content": "## Abstract\n\nAbstract Language models are aligned to emulate the collective voice of many, resulting in outputs that align with no one in particular. Steering LLMs away from generic output is possible through supervised finetuning or RLHF, but requires prohibitively large datasets for new ad-hoc tasks... | [
{
"id": "90TTdZLDNb",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces a novel method, Demonstration Iterated Task Optimization (DITTO), for training large language models (LLMs) with expert demonstration dataset... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;3;2",
"soundness_avg": 2.75,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"presentation": "3;4;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.648268"
} | {
"id": "auLYy3KyRl",
"metareview": "## Summary\nThis paper introduces Demonstration ITerated Task Optimization (DITTO), a method for aligning language models to specific tasks using fewer than 16 demonstrations. Unlike methods like RLHF or supervised fine-tuning, which often require large datasets, DITTO leverages... | {
"decision": "Accept (Poster)"
} |
1qq1QJKM5q | 2410.08003v2 | More Experts Than Galaxies: Conditionally-Overlapping Experts with Biologically-Inspired Fixed Routing | {
"content": "## Abstract\n\nAbstract The evolution of biological neural systems has led to both modularity\nand sparse coding, which\nenables efficiency in energy usage, and robustness across the diversity of tasks in the lifespan. In contrast, standard neural networks rely on dense, non-specialized architectures, w... | [
{
"id": "TyBUL3aCZL",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces Conditionally Overlapping Mixture of ExperTs (COMET).\nCOMET uses biologically inspired, fixed random projections to generate binary masks th... | {
"rating": "3;6;6",
"rating_avg": 5,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;3;2",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.649113"
} | {
"id": "UDGPZ7JQ2u",
"metareview": "**Summary**\n\nThe paper introduces Conditionally Overlapping Mixture of Experts (COMET), a novel method for enhancing sparse neural networks by using biologically inspired, fixed random projections to generate input-dependent binary masks. These masks define subnetworks called ... | {
"decision": "Accept (Poster)"
} |
1t1YSuBv3T | 2408.15037v1 | Evidence-Enhanced Triplet Generation Framework for Hallucination Alleviation in Generative Question Answering | {
"content": "## Abstract\n\nAbstract To\naddress the hallucination in generative question answering (GQA) where the answer can not be derived from the document, we propose a novel evidence-enhanced triplet generation framework,\nEATQA, encouraging the model to\npredict all the combinations of ⟨Question, Evidence, An... | [
{
"id": "D7kX45mFTN",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes EATQA (Evidence-Enhanced Triplet Generation Framework), designed to reduce hallucinations in Generative Question Answering (GQA). EATQA leverag... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;3;3",
"confidence_avg": 3.3333333333333335,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;3;2",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;3",
"presentation_avg": 2.33333333333333... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.649790"
} | {
"id": "lfTzv36UJT",
"metareview": "The paper introduces a method designed to alleviate hallucinations in Generative Question Answering by employing a structured approach. This involves the creation of triplets consisting of a Question, Evidence, and Answer, which are used to enhance logical consistency.\n\nWhile ... | {
"decision": "Reject"
} |
1tBvzOYTLF | 2410.05193v1 | RevisEval: Improving LLM-as-a-Judge via Response-Adapted References | {
"content": "## Abstract\n\nAbstract With significant efforts in recent studies, LLM-as-a-Judge has become a cost-effective alternative to human evaluation for assessing the text generation quality in a wide range of tasks.\nHowever, there still remains a reliability gap between LLM-as-a-Judge and human evaluation. ... | [
{
"id": "JgZog0JBKW",
"initial_rating": 5,
"confidence": 5,
"soundness": 1,
"contribution": 3,
"presentation": 4,
"summary": "The paper proposes an interesting method “RevisEval” which explores a new approach to performing reference-based evaluation by modifying references based on the r... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "4;4;5;4",
"confidence_avg": 4.25,
"soundness": "2;4;1;4",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;3;4;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.650451"
} | {
"id": "MDehs40287",
"metareview": "The paper proposes RevisEval which improves on the standard LLM-as-a-judge pipeline by modifying the reference response to align better with the generated output. Reviewers generally believe the idea has promise and it improves over the baseline LLM-as-a-judge methods. They do r... | {
"decision": "Accept (Poster)"
} |
1uLW9eYNJB | 2410.00938v1 | MoS: Unleashing Parameter Efficiency of Low-Rank Adaptation with Mixture of Shards | {
"content": "## Abstract\n\nAbstract The rapid scaling of large language models necessitates more lightweight finetuning methods to reduce the explosive GPU memory overhead when numerous customized models are served simultaneously.\nTargeting more parameter-efficient low-rank adaptation (LoRA), parameter sharing pre... | [
{
"id": "gMWR9MSvhk",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "The paper investigates a more lightweight solution than LoRA in order to serve a large number of finetuned models at the same time. Based on a finding that excess... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;4;5;3",
"confidence_avg": 3.75,
"soundness": "2;3;4;4",
"soundness_avg": 3.25,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.651243"
} | {
"id": "FuBuRStwXc",
"metareview": "The paper presents a novel method called Mixture of Shards that combines inter-layer and intra-layer sharing with lightweight differentiation strategies to enhance parameter efficiency and model performance for large models. The MoS method outperforms existing approaches like Lo... | {
"decision": "Accept (Poster)"
} |
1v7SRWsYve | 2406.07529v4 | MAP: Low-compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation | {
"content": "## Abstract\n\nAbstract Model merging has emerged as an effective approach to combine multiple single-task models into a multitask model. This process typically involves computing a weighted average of the model parameters without any additional training. Existing model-merging methods focus on enhancin... | [
{
"id": "YNhjcWtIWI",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "When merging models (generally finetuned on different tasks), many techniques boil down to a weighted sum (generally of \"task vectors\", the difference between t... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;2;2",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.652124"
} | {
"id": "LYFAc5fxXc",
"metareview": "The paper looks at model merging for creating a model to perform on known tasks. \nCurrent merging methods often propose a single merged models, this paper proposes a new method to create multiple models that have different capabilities as shown by tradeoffs between which evalu... | {
"decision": "Accept (Poster)"
} |
1vrpdV9U3i | 2409.06142v2 | Variational Search Distributions | {
"content": "## Abstract\n\nAbstract We develop variational search distributions (VSD) , a method for finding discrete, combinatorial designs of a rare desired class in a batch sequential manner with a fixed experimental budget. We formalize the requirements and desiderata for this problem and formulate a solution v... | [
{
"id": "Ha686hWof3",
"initial_rating": 3,
"confidence": 5,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "This paper casts sequential black-box optimization as a variational inference (i.e. amortized optimization) problem, and uses this perspective to unify a collecti... | {
"rating": "3;6;6;6",
"rating_avg": 5.25,
"confidence": "5;3;2;3",
"confidence_avg": 3.25,
"soundness": "4;4;3;3",
"soundness_avg": 3.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;2;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.653134"
} | {
"id": "xwMStviKDi",
"metareview": "This paper considers an active search problem and presents a method to generate new designs of rare desired class under budget constraints. The target problem is of discrete and combinatorial nature. The proposed algorithm is based on variational inference. Theoretical analysis ... | {
"decision": "Accept (Poster)"
} |
1x1gGg49jr | 2410.11682v1 | SurFhead: Affine Rig Blending for Geometrically Accurate 2D Gaussian Surfel Head Avatars | {
"content": "## Abstract\n\nAbstract Recent advancements in head avatar rendering using Gaussian primitives have achieved significantly high-fidelity results. Although precise head geometry is crucial for applications like mesh reconstruction and relighting, current methods struggle to capture intricate geometric de... | [
{
"id": "pxO9S7LfeZ",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a method for learning head avatars based on 2D Gaussian splatting. To make the Gaussian surfels better handle the stretch and shear deformatio... | {
"rating": "6;6;6;6",
"rating_avg": 6,
"confidence": "3;2;5;4",
"confidence_avg": 3.5,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.654072"
} | {
"id": "oChazTWaUG",
"metareview": "The paper introduces a new model within the Gaussian Splatting framework designed to capture realistic and detailed head deformations from RGB videos. The model enhances geometric accuracy and detail, particularly in complex scenarios like sharp reflections and exaggerated defor... | {
"decision": "Accept (Poster)"
} |
1xG3MN1RRW | 2410.04417v2 | SparseVLM: Visual Token Sparsification for Efficient Vision Language Models Inference | {
"content": "## Abstract\n\nAbstract In vision-language models (VLMs), visual tokens usually consume a significant amount of computational overhead, despite their sparser information density compared to text tokens. To address this, most existing methods learn a network to prune redundant visual tokens and require a... | [
{
"id": "x8ANCrIRbE",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces SparseVLM, a method to accelerate vision-language models (VLMs) by prunning vision tokens incrementally over layers, based on its significanc... | {
"rating": "3;5;5;6;6",
"rating_avg": 5,
"confidence": "5;4;5;5;4",
"confidence_avg": 4.6,
"soundness": "2;3;2;3;2",
"soundness_avg": 2.4,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "2;3;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.654727"
} | {
"id": "qQ30t7aqKI",
"metareview": "The authors address a timely and relevant problem: The number of visual tokens to a VLM are often redundant and substantially increase the computational cost of the model. The authors propose to address this by proposing an iterative token sparification strategy that reduces the... | {
"decision": "Reject"
} |
1xzqz73hvL | 2410.18837v1 | High-dimensional Analysis of Knowledge Distillation: Weak-to-Strong Generalization and Scaling Laws | {
"content": "## Abstract\n\nAbstract A growing number of machine learning scenarios rely on knowledge distillation where one uses the output of a surrogate model as labels to supervise the training of a target model. In this work, we provide a sharp characterization of this process for ridgeless, high-dimensional re... | [
{
"id": "UDmNf2ZgZ1",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "In this paper, the authors propose a precise characterization of the benefits of knowledge distillation, mostly in the context of gaussian linear regression. In p... | {
"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": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.655659"
} | {
"id": "SnhD1FbBGg",
"metareview": "The work presents new results on knowledge distillation, with focus on the linear Gaussian setting. The work characterizes excess risk of such linear regression based on pseudo labels from a surrogate model, based on covariate shift. The work presents a clear set of results, de... | {
"decision": "Accept (Spotlight)"
} |
1yJ3IDpb1D | 2406.14341v2 | HoTPP Benchmark: Are We Good at the Long Horizon Events Forecasting? | {
"content": "## Abstract\n\nAbstract Accurately forecasting multiple future events within a given time horizon is crucial for finance, retail, social networks, and healthcare applications. Event timing and labels are typically modeled using Marked Temporal Point Processes (MTPP), with evaluations often focused on ne... | [
{
"id": "PHVj1BZc9C",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces HoTPP (Horizon Temporal Point Process Benchmark), a benchmark for evaluating long-horizon event sequence prediction models. Its main contrib... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "2;4;2;3",
"confidence_avg": 2.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "2;2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.656678"
} | {
"id": "TibQZ0ycQH",
"metareview": "The reviewers find the paper difficult to understand and raise many general questions that cannot be resolved in a rebuttal as most of them require substantial changes to the original submission and in sum a major revision. For example, it does not become obvious why the solved... | {
"decision": "Reject"
} |
1yJP5TVWih | 2410.10609v2 | Lambda-Skip Connections: the architectural component that prevents Rank Collapse | {
"content": "## Abstract\n\nAbstract Rank collapse, a phenomenon where embedding vectors in sequence models rapidly converge to a uniform token or equilibrium state, has recently gained attention in the deep learning literature. This phenomenon leads to reduced expressivity and potential training instabilities due t... | [
{
"id": "JuON1TuK6P",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "Dao and Gu [https://arxiv.org/pdf/2405.21060] established a form of equivalence between transformers and continuous-time state-space models. In a different devel... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;3;2;4",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"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": "2026-01-14T22:16:00.657467"
} | {
"id": "kXPdpHMmWX",
"metareview": "The paper discusses the phenomenon of rank collapse in LLMs, extending the study done on transformers to the SSM architecture. This issue is highly motivated given the popularity of SSM-based LLMs. The reviews note that the paper provides valuable insights to the problem coupled... | {
"decision": "Accept (Poster)"
} |
204sPiwBbB | 2410.16509v1 | Learning from others' mistakes: Finetuning machine translation models with span-level error annotations | {
"content": "## Abstract\n\nAbstract Despite growing interest in incorporating feedback to improve language models, most efforts focus only on sequence-level annotations.\nIn this work, we explore the potential of utilizing fine-grained span-level annotations from offline datasets to improve model quality.\nWe devel... | [
{
"id": "OoGCpZwStM",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This work proposes a new method called Training with Annotations (TWA) that leverages the MT evaluation annotation data to improve the quality of machine translat... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "3;3;5;4",
"confidence_avg": 3.75,
"soundness": "2;2;2;4",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;2;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.658202"
} | {
"id": "DV5qOW0kOS",
"metareview": "This paper introduces a novel method, Training with Annotations (TWA), that leverages machine quality measurement (MQM) annotation data to improve machine translation systems. Specifically, it uses span-level error annotations to create fine-grained supervision via an additional... | {
"decision": "Reject"
} |
20mMK8UlFh | 2410.01944v1 | One-step Noisy Label Mitigation | {
"content": "## Abstract\n\nAbstract Mitigating the detrimental effects of noisy labels on the training process has become increasingly critical, as obtaining entirely clean or human-annotated samples for large-scale pre-training tasks is often impractical. Nonetheless, existing noise mitigation methods often encoun... | [
{
"id": "Oo3cKVIUhY",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper addresses the challenge of mitigating the detrimental effects of noisy labels in large-scale pre-training tasks, where obtaining entirely clean data is ... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "3;5;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;1;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": "2026-01-14T22:16:00.658850"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
20qZK2T7fa | 2410.07994v1 | Neuroplastic Expansion in Deep Reinforcement Learning | {
"content": "## Abstract\n\nAbstract The loss of plasticity in learning agents, analogous to the solidification of neural pathways in biological brains, significantly impedes learning and adaptation in reinforcement learning due to its non-stationary nature. To address this fundamental challenge, we propose a novel ... | [
{
"id": "ewwnHMbTZZ",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a novel idea, Neuroplastic Expansion (NE), to address the problem of plasticity loss in reinforcement learning (RL). The paper is well-writt... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;5;5;3",
"confidence_avg": 4.25,
"soundness": "2;2;3;2",
"soundness_avg": 2.25,
"contribution": "2;3;3;2",
"contribution_avg": 2.5,
"presentation": "1;2;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.659543"
} | {
"id": "VFBz84sT1g",
"metareview": "This paper addresses the stability-plasticity dilemma inspired by dormant neuron pruning and expansion of connection topology inspired by cortical expansion in cognitive science. The approach is intuitive, but the actual details of expansion seem quite complicated. An additional... | {
"decision": "Accept (Poster)"
} |
23uY3FpQxc | 2410.03435v1 | A General Framework for Producing Interpretable Semantic Text Embeddings | {
"content": "## Abstract\n\nAbstract Semantic text embedding is essential to many tasks in Natural Language Processing (NLP).\nWhile black-box models are capable of generating high-quality embeddings, their lack of interpretability limits their use in tasks that demand transparency.\nRecent approaches have improved ... | [
{
"id": "3CV2z3hhzm",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes CQG-MVQA, an intepretable embedding framework. The framework generates questions and binary answers about texts, and trains binary classifiers ... | {
"rating": "5;5;6;6;6",
"rating_avg": 5.6,
"confidence": "4;4;3;4;3",
"confidence_avg": 3.6,
"soundness": "3;2;3;4;3",
"soundness_avg": 3,
"contribution": "3;2;3;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": "2026-01-14T22:16:00.660373"
} | {
"id": "7gcqHX86xb",
"metareview": "This paper introduces a framework called CQG-MBQA to create interpretable text embeddings, which automatically produces yes/no questions that capture the differences between texts. \n\nThe reviewers recognized the authors’ contributions to their proposed framework for generating... | {
"decision": "Accept (Poster)"
} |
25kAzqzTrz | 2410.11206v1 | Towards Understanding Why FixMatch Generalizes Better Than Supervised Learning | {
"content": "## Abstract\n\nAbstract Semi-supervised learning (SSL), exemplified by FixMatch (Sohn et al., 2020 ) , has shown significant generalization advantages over supervised learning (SL), particularly in the context of deep neural networks (DNNs). However, it is still unclear, from a theoretical standpoint, w... | [
{
"id": "sAISYtcHgM",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "The paper studies the feature learning process of neural networks trained with the FixMatch method, which is a semi-supervised learning method, demonstrating its ... | {
"rating": "3;6;8;8",
"rating_avg": 6.25,
"confidence": "3;4;4;3",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "1;3;4;4",
"contribution_avg": 3,
"presentation": "4;3;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.661750"
} | {
"id": "AFAYMJmlEP",
"metareview": "This work theoretically justifies why FixMatch-like self-supervised learning methods outperform supervised learning (SL) in generalization for deep networks, showing that FixMatch learns all class features while SL captures only a subset. The authors introduce SA-FixMatch, an en... | {
"decision": "Accept (Oral)"
} |
25l4SWH2eS | 2409.08240v3 | IFAdapter: Instance feature control for grounded Text-to-Image Generation | {
"content": "## Abstract\n\nAbstract While Text-to-Image (T2I) diffusion models excel at generating visually appealing images of individual instances, they struggle to accurately position and control the features generation of multiple instances. The Layout-to-Image (L2I) task was introduced to address the positioni... | [
{
"id": "bTE3BBslHK",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper aims to improve the capability of Text-to-Image diffusion models in generating precise features and positioning multiple instances in images. The propo... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "5;4;4;5",
"confidence_avg": 4.5,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"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": "2026-01-14T22:16:00.662905"
} | {
"id": "qmJdGoIDTi",
"metareview": "Summary\nThis paper studies the problem of controllable image generation where the scene layout can be specified using a global caption, instance locations, and instance specific captions. The authors propose an adapter based technique called IFA that fuses such instance conditi... | {
"decision": "Reject"
} |
26kgSlMmhA | 2407.09357v2 | Any-Property-Conditional Molecule Generation with Self-Criticism using Spanning Trees | {
"content": "## Abstract\n\nAbstract Generating novel molecules is challenging, with most representations leading to generative models producing many invalid molecules. Spanning Tree-based Graph Generation (STGG) ( ahn2021spanning ) is a promising approach to ensure the generation of valid molecules, outperforming s... | [
{
"id": "QCklDuSr8V",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper presents STGG+, a model for molecule generation with conditional property control. The model also has the ability of self-criticism to select optimal ou... | {
"rating": "3;3;5;5;5;5;6",
"rating_avg": 4.571428571428571,
"confidence": "3;4;4;2;4;3;2",
"confidence_avg": 3.142857142857143,
"soundness": "3;2;3;3;2;3;3",
"soundness_avg": 2.7142857142857144,
"contribution": "3;1;2;3;2;2;3",
"contribution_avg": 2.2857142857142856,
"presentation": "2;2;4;4;3;2;3",... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.663463"
} | {
"id": "ntUbS58wCe",
"metareview": "**Summary**\n\nThis work extends an unconditional molecular generation method ‘Spanning Tree-based Graph Generation’ (STGG) to enable conditional generation given some desired properties. Toward that end, the authors enhance the Transformer architecture of STGG to include (a) ra... | {
"decision": "Reject"
} |
26oSbRRpEY | 2403.14773v1 | StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text | {
"content": "## Abstract\n\nAbstract Text-to-video diffusion models enable the generation of high-quality videos that follow text instructions, making it easy to create diverse and individual content.\nHowever, existing approaches mostly focus on high-quality short video generation (typically 16 or 24 frames), endin... | [
{
"id": "xhy1HRpCw6",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents StreamingT2V, a method for generating high-quality, extended videos from text prompts, specifically addressing the challenge of ensuring smoot... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "5;3;4;5",
"confidence_avg": 4.25,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;2;2",
"contribution_avg": 2.25,
"presentation": "2;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": "2026-01-14T22:16:00.664364"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
27Qk18IZum | 2409.06316v1 | PharmacoMatch: Efficient 3D Pharmacophore Screening via Neural Subgraph Matching | {
"content": "## Abstract\n\nAbstract The increasing size of screening libraries poses a significant challenge for the development of virtual screening methods for drug discovery, necessitating a re-evaluation of traditional approaches in the era of big data.\nAlthough 3D pharmacophore screening remains a prevalent t... | [
{
"id": "y1yq3iZ1lI",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a novel contrastive learning approach based on neural subgraph matching, i.e., PharmacoMatch, and the authors claim that it reinterprets pha... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "3;4;3;4",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.665153"
} | {
"id": "FDFwlDqsug",
"metareview": "This paper introduces a new virtual screening method based on contrastive learning via neural subgraph matching. Of particular relevance for me is that the method is substantially _faster_ than existing methods; this is an aspect of machine-learning models that we typically do n... | {
"decision": "Accept (Poster)"
} |
27n0kvWgqT | 2410.09016v1 | Parameter-Efficient Fine-Tuning of State Space Models | {
"content": "## Abstract\n\nAbstract Deep State Space Models (SSMs), such as Mamba (Gu & Dao, 2024 ) , have emerged as powerful tools for language modeling, offering high performance with efficient inference and linear scaling in sequence length.\nHowever, the application of parameter-efficient fine-tuning (PEFT) me... | [
{
"id": "s4G1d02ZNt",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents the first study on the performance of PEFT methods applied to SSM-based models. Specifically, prompt-based and parameter-based methods are inv... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"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": "2026-01-14T22:16:00.666283"
} | {
"id": "xGrBfPZE2c",
"metareview": "This paper presents a systematic investigation into parameter-efficient fine-tuning (PEFT) methods for Deep State Space Models (SSMs), with a particular focus on language modeling tasks. They introduce SDLoRA, a novel approach that selectively updates certain channels and states... | {
"decision": "Reject"
} |
28U5Olm32r | 2410.06851v1 | Understanding Model Ensemble in Transferable Adversarial Attack | {
"content": "## Abstract\n\nAbstract {NoHyper} † † ⋆ ⋆ \\star ⋆ indicates equal contribution. {NoHyper} † † † † \\dagger † indicates the corresponding author. Model ensemble adversarial attack has become a powerful method for generating transferable adversarial examples that can target even unknown models, but its t... | [
{
"id": "M2Bl7GeX7b",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This submission theoretically studies the transferability error - the chance of being successful if the attack is generated by transferring from an ensemble of mo... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;4;2;3",
"soundness_avg": 2.75,
"contribution": "2;4;2;3",
"contribution_avg": 2.75,
"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": "2026-01-14T22:16:00.667370"
} | {
"id": "z9gSNiUk1w",
"metareview": "This paper introduces a theoretical framework for understanding transferable adversarial attacks, proposing concepts such as transferability error and a vulnerability-diversity decomposition, with practical guidelines supported by experiments on standard datasets. While the theo... | {
"decision": "Reject"
} |
28abpUEICJ | 2405.17395v1 | CREIMBO: Cross-Regional Ensemble Interactions in Multi-view Brain Observations | {
"content": "## Abstract\n\nAbstract Modern recordings of neural activity provide diverse observations of neurons across brain areas, behavioral conditions, and subjects—thus presenting an exciting opportunity to reveal the fundamentals of brain-wide dynamics underlying cognitive function.\nCurrent analysis methods,... | [
{
"id": "6F0lzFC8OJ",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "Through the study, the authors have proposed a novel analytical approach (named as “CREIMBO”) for learning dynamics of latent representations from high-dimensiona... | {
"rating": "3;6;8",
"rating_avg": 5.666666666666667,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "3;3;4",
"soundness_avg": 3.3333333333333335,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "4;3;3",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.668149"
} | {
"id": "T3VJ8Orw8P",
"metareview": "The authors propose CREIMBO, a novel model that extracts interpretable neural subcircuits from multi-region, multi-session electrophysiological data, maintaining both compression and interpretability. CREIMBO demonstrates its effectiveness in both simulated and real data, reveal... | {
"decision": "Accept (Spotlight)"
} |
28qOQwjuma | 2410.10083v2 | Beyond Graphs: Can Large Language Models Comprehend Hypergraphs? | {
"content": "## Abstract\n\nAbstract Existing benchmarks like NLGraph and GraphQA evaluate LLMs on graphs by focusing mainly on pairwise relationships, overlooking the high-order correlations found in real-world data. Hypergraphs, which can model complex beyond-pairwise relationships, offer a more robust framework b... | [
{
"id": "eM2xRic7kh",
"initial_rating": 8,
"confidence": 5,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces LLM4Hypergraph, a benchmark designed to evaluate large language models' (LLMs) understanding of hypergraphs, which can capture complex, multi... | {
"rating": "3;5;8",
"rating_avg": 5.333333333333333,
"confidence": "4;4;5",
"confidence_avg": 4.333333333333333,
"soundness": "2;2;4",
"soundness_avg": 2.6666666666666665,
"contribution": "1;3;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;3",
"presentation_avg": 2.666666666666666... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.669002"
} | {
"id": "UNkNOEkXsA",
"metareview": "The authors present the first comprehensive benchmark for LLM evaluation on hypergraph reasoning. This benchmark covers eight tasks, evaluates six LLMs, and incorporates both synthetic and real-world datasets. Furthermore, the authors propose novel prompt engineering techniques ... | {
"decision": "Accept (Poster)"
} |
29JDZxRgPZ | 2405.17366v1 | EM-GANSim: Real-time and Accurate EM Simulation Using Conditional GANs for 3D Indoor Scenes | {
"content": "## Abstract\n\nAbstract We present a novel machine-learning (ML) approach (EM-GANSim) for real-time electromagnetic (EM) propagation that is used for wireless communication simulation in 3D indoor environments. Our approach uses a modified conditional Generative Adversarial Network (GAN) that incorporat... | [
{
"id": "XS14WKbfXq",
"initial_rating": 5,
"confidence": 2,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The paper presents a generative framework for simulating Electro-Magnetic wave propagation, as a faster replacement for ray tracing approaches usually used in thi... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "3;2;2;3",
"confidence_avg": 2.5,
"soundness": "3;2;3;4",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"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": "2026-01-14T22:16:00.669651"
} | {
"id": "P5ghEjMfCS",
"metareview": "The paper proposes an approach to learning a real-time-capable EM simulation via conditional GANs. It proposes to incorporate physical constraints via the loss function to improve the accuracy of predicted power distributions. The paper further presents a synthetic dataset of 3D... | {
"decision": "Reject"
} |
29LC48aY3U | 2410.14425v1 | Backdoor Attacks for LLMs with Weak-To-Strong Knowledge Distillation | {
"content": "## Abstract\n\nAbstract Parameter-efficient fine-tuning (PEFT) can bridge the gap between large language models (LLMs) and downstream tasks. However, PEFT has been proven vulnerable to malicious attacks. Research indicates that poisoned LLMs, even after PEFT, retain the capability to activate internaliz... | [
{
"id": "MH6KauHAy7",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces W2SAttack, a method for injecting clean-label backdoors into LLMs. The approach stems from the observation that successfully injecting clean... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "5;4;3;3",
"confidence_avg": 3.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.670282"
} | {
"id": "hWPwObQsk2",
"metareview": "The paper proposes a new backdoor attack in the clean label setting.\nAs noted by reviewer LsBW, there seems to be an inconsistency in the considered threat model, where the attacker has full control over the training process (which is standard in some backdoor works), but still... | {
"decision": "Reject"
} |
2AWZTv6kgV | 2410.23667v1 | Projected Neural Differential Equations for Learning Constrained Dynamics | {
"content": "## Abstract\n\nAbstract Neural differential equations offer a powerful approach for learning dynamics from data.\nHowever, they do not impose known constraints that should be obeyed by the learned model.\nIt is well-known that enforcing constraints in surrogate models can enhance their generalizability ... | [
{
"id": "PKMqyhDGdZ",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "The paper addresses the challenge of learning constrained dynamical systems in the context of neural differential equations (i.e., NDEs herafter). This term of ND... | {
"rating": "1;5;5;8",
"rating_avg": 4.75,
"confidence": "5;4;3;4",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "1;3;1;3",
"contribution_avg": 2,
"presentation": "2;4;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.671354"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
2Akf4BBCKo | 2410.18517v1 | KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing | {
"content": "## Abstract\n\nAbstract The development of large language models (LLMs) has significantly expanded model sizes, resulting in substantial GPU memory requirements during inference. The key and value storage of the attention map in the KV (key-value) cache accounts for more than 80% of this memory consumpt... | [
{
"id": "dbv5UzVAWL",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a new inter-layer KV cache compression technique through layer-wise KV cache dis-similarity search and sharing. The layers are ranked pairwi... | {
"rating": "3;3;3;5;5",
"rating_avg": 3.8,
"confidence": "4;4;4;3;4",
"confidence_avg": 3.8,
"soundness": "1;2;1;3;2",
"soundness_avg": 1.8,
"contribution": "1;2;2;3;3",
"contribution_avg": 2.2,
"presentation": "2;3;3;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": "2026-01-14T22:16:00.672036"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
2DD4AXOAZ8 | 2409.15012v1 | Inference-Friendly Models With MixAttention | {
"content": "## Abstract\n\nAbstract The size of the key-value (KV) cache plays a critical role in determining both the maximum context length and the number of concurrent requests supported during inference in modern language models. The KV cache size grows proportionally with the number of attention heads and the ... | [
{
"id": "xJWGVi2elR",
"initial_rating": 1,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 1,
"summary": "The authors introduce MixAttention, an architecture that employs sliding window attention to store only recent tokens while sharing KV caches across layers. They ... | {
"rating": "1;1;3;3",
"rating_avg": 2,
"confidence": "5;4;4;5",
"confidence_avg": 4.5,
"soundness": "3;2;2;2",
"soundness_avg": 2.25,
"contribution": "1;1;2;2",
"contribution_avg": 1.5,
"presentation": "2;1;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.672605"
} | {
"id": "fRVDxsJlVH",
"metareview": "The paper proposes MixAttention, an architectural modification for language models that combines sliding window attention with KV cache sharing across layers to reduce memory usage and improve inference speed.\n\nThe main concerns were the lack of novelty as it primarily reprodu... | {
"decision": "Reject"
} |
2ET561DyPe | 2411.01099v1 | Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement | {
"content": "## Abstract\n\nAbstract We propose Few-Class Arena ( FCA ), as a unified benchmark with focus on testing efficient image classification models for few classes. A wide variety of benchmark datasets with many classes (80-1000) have been created to assist Computer Vision architectural evolution. An increas... | [
{
"id": "sSnbk0hmtU",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper tackles the problem of choosing image classifiers for tasks with only a small number of categories (\"Few-Class\"). To do so, they introduce a new benc... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.673211"
} | {
"id": "RCWcYSEqDZ",
"metareview": "This work aims to benchmark image classification in the few-class regime as a proxy for performance on real-world tasks of this size. Such a focus is complementary to existing large-scale benchmarks such as ImageNet, COCO, etc. with 10s of classes or 1000 classes. The contributi... | {
"decision": "Accept (Poster)"
} |
2FMdrDp3zI | 2410.12537v1 | Is Complex Query Answering Really Complex? | {
"content": "## Abstract\n\nAbstract Complex query answering (CQA) on knowledge graphs (KGs) is gaining momentum as a challenging reasoning task.\nIn this paper, we show that the current benchmarks for CQA are not really complex , and the way they are built distorts our perception of progress in this field.\nFor exa... | [
{
"id": "BX57pgyhqn",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 4,
"summary": "This manuscript presents a data-level study of knowledge graph complex query answering. The main argument is that the query-target pairs in existing datasets (q, ... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;1",
"contribution_avg": 1.75,
"presentation": "4;3;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.673881"
} | {
"id": "Vz9c8Zm7CJ",
"metareview": "This paper shows that current benchmarks for CQA are not really complex, and that in these benchmarks most queries (up to 98% for some query types) can be reduced to simpler problems, e.g., link prediction, where only one link needs to be predicted. The performance of state-of-t... | {
"decision": "Reject"
} |
2G021ZqUEZ | 2410.11843v1 | From Commands to Prompts: LLM-based Semantic File System | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have demonstrated significant potential in the development of intelligent applications and systems such as LLM-based agents and agent operating systems (AIOS). However, when these applications and systems interact with the underlying file system, the ... | [
{
"id": "78AiPQPGPx",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces an LLM-based Semantic File System (LSFS), designed to improve file management through natural language prompts, rather than traditional comm... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.674583"
} | {
"id": "JkvS0r6DEk",
"metareview": "The paper proposes LSFS (LLM-based Semantic File System), a novel approach that enhances traditional file systems with semantic understanding through LLMs. It enables natural language interactions for file operations with some designed safety mechanisms. The authors demonstrate ... | {
"decision": "Accept (Poster)"
} |
2GcR9bO620 | 2411.00121v1 | I Can Hear You: Selective Robust Training for Deepfake Audio Detection | {
"content": "## Abstract\n\nAbstract Recent advances in AI-generated voices have intensified the challenge of detecting deepfake audio, posing further risks for the spread of scams and disinformation. To tackle this issue, we establish the largest public voice dataset to date, named DeepFakeVox-HQ , comprising 1.3 m... | [
{
"id": "DoeAWiGmLw",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The paper attempts to improve deepfake detection by (1) proposing a large training and evaluation dataset called DeepFakeVox-HQ containing diverse synthetic and r... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "3;3;5;4",
"confidence_avg": 3.75,
"soundness": "4;3;3;2",
"soundness_avg": 3,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "4;2;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.675251"
} | {
"id": "kd5AP2W4gt",
"metareview": "This paper makes significant advancements in deepfake detection through three key contributions: (1) Introduction of DeepFakeVox-HQ: A comprehensive dataset comprising high-quality synthetic and real speech recordings, designed to benefit the research community. (2) Development ... | {
"decision": "Accept (Poster)"
} |
2H6KhX1kJr | 2405.20180v1 | Transformers and slot encoding for sample efficient physical world modelling | {
"content": "## Abstract\n\nAbstract World modelling, i.e. building a representation of the rules that govern the world so as to predict its evolution, is an essential ability for any agent interacting with the physical world.\nRecent applications of the Transformer architecture to the problem of world modelling fro... | [
{
"id": "sDXtQbfFjV",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper present a slotted recurrent network model which uses transformers as the main backbone for \"world modeling\". In this context the resulting model is a... | {
"rating": "3;3;3;3",
"rating_avg": 3,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;1;2",
"soundness_avg": 2,
"contribution": "1;2;1;2",
"contribution_avg": 1.5,
"presentation": "1;3;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.675924"
} | {
"id": "8hXTBqGum1",
"metareview": "This paper introduces a framework for world modelling using tokenised (object-centric) representations. A key part of the method is the combination of a transformer architecture with a slot-attention-like mechanism. Experiments are run on the PHYRE simulated physical reasoning b... | {
"decision": "Reject"
} |
2IBdk8cUdC | 2406.05985v2 | Topo-Field: Topometric mapping with Brain-inspired Hierarchical Layout-Object-Position Fields | {
"content": "## Abstract\n\nAbstract Spatial cognition empowers animals with remarkably efficient navigation abilities, largely depending on the scene-level understanding of spatial environments. Recently, it has been found that a neural population in the postrhinal cortex of rat brains is more strongly tuned to the... | [
{
"id": "n9QgfNubNq",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "The paper introduces Topo-Field, a framework designed to enhance mobile robot navigation by integrating detailed semantic information about layouts, objects, and ... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "5;3;3;4",
"confidence_avg": 3.75,
"soundness": "1;2;2;2",
"soundness_avg": 1.75,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "1;3;2;1",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.676860"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
2IhkyiF3to | 2411.00147v1 | Mutual Information Preserving Neural Network Pruning | {
"content": "## Abstract\n\nAbstract Model pruning is attracting increasing interest because of its positive implications in terms of resource consumption and costs. A variety of methods have been developed in the past years. In particular, structured pruning techniques discern the importance of nodes in neural netw... | [
{
"id": "whjGqjXr7G",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The authors propose MIPP to enable real-time pruning, whole-layer pruning and global re-training guarantees for improving the performance of network pruning. Thro... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;2;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.677734"
} | {
"id": "cQ2fCtnO7L",
"metareview": "The submission proposes Mutual Information Preserving Pruning (MIPP), a method to prune filters/nodes in neural networks that aims to preserve the mutual information between adjacent layers in a neural network.\nAfter the inital round of reviews, this submission received scores ... | {
"decision": "Reject"
} |
2J18i8T0oI | 2410.06672v2 | Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures | {
"content": "## Abstract\n\nAbstract The hypothesis of Universality in interpretability suggests that different neural networks may converge to\nimplement similar algorithms on similar tasks. In this work, we investigate two mainstream architectures\nfor language modeling, namely Transformers and Mambas, to explore ... | [
{
"id": "WDVqjcl7JG",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper investigates the \"universality hypothesis\" in mechanistic interpretability, which suggests that different neural network architectures may converge to... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "3;2;3;4",
"soundness_avg": 3,
"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": "2026-01-14T22:16:00.678464"
} | {
"id": "lJnu0CioDF",
"metareview": "The paper investigates the universality hypothesis, showing that Transformers and Mambas, despite architectural differences, learn similar features for language modeling tasks. Using Sparse Autoencoders (SAEs), the authors demonstrate cross-architecture feature similarity and id... | {
"decision": "Accept (Poster)"
} |
2L1OxhQCwS | 2309.11400v1 | Transformers versus LSTMs for electronic trading | {
"content": "## Abstract\n\nAbstract With the rapid development of artificial intelligence, long short term memory (LSTM), one kind of recurrent neural network (RNN), has been widely applied in time series prediction. Like RNN, Transformer is designed to handle the sequential data. As Transformer achieved great succ... | [
{
"id": "flDuAZNkXt",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 1,
"presentation": 2,
"summary": "This research compares the effectiveness of Transformer and LSTM architectures in financial forecasting. The study examines both model types using high-frequency ... | {
"rating": "3;3;3;3;3;5",
"rating_avg": 3.3333333333333335,
"confidence": "5;3;4;4;3;3",
"confidence_avg": 3.6666666666666665,
"soundness": "1;1;2;3;2;3",
"soundness_avg": 2,
"contribution": "1;1;2;2;1;3",
"contribution_avg": 1.6666666666666667,
"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": "2026-01-14T22:16:00.679192"
} | {
"id": "Sc57fKUlrO",
"metareview": "A. Scientific Claims and Findings:\n\nThe paper compares the effectiveness of Transformer and LSTM architectures for financial forecasting using high-frequency trading data. The authors introduce a new LSTM-based model called DLSTM and a finance-specific Transformer. Their resul... | {
"decision": "Reject"
} |
2MqyCIxLSi | 2410.06530v2 | TopoTune: A Framework for Generalized Combinatorial Complex Neural Networks | {
"content": "## Abstract\n\nAbstract Graph Neural Networks (GNNs) excel in learning from relational datasets, processing node and edge features in a way that preserves the symmetries of the graph domain. However, many complex systems—such as biological or social networks—involve multiway complex interactions that ar... | [
{
"id": "wtvlQgY9q9",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper focuses on the topological deep learning (TDL) models in particular CCNNs and proposes a new powerful graph-based methodology for new TDL architectures,... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.680080"
} | {
"id": "snTrNK3z6c",
"metareview": "This paper proposes a new framework for topological deep learning that addresses several open problems in a recent position paper.\nAll reviewers agree this is a very valuable contribution, and the paper is well-structured and well-written. \nHowever, during the discussion perio... | {
"decision": "Reject"
} |
2NqssmiXLu | 2410.15756v1 | Automated Proof Generation for Rust Code via Self-Evolution | {
"content": "## Abstract\n\nAbstract Ensuring correctness is crucial for code generation.\nFormal verification offers a definitive assurance of correctness, but\ndemands substantial human effort in proof construction and hence raises a pressing need for automation. The primary obstacle lies in the severe lack of dat... | [
{
"id": "uz9z3Wghvd",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper proposes SAFE, a data generation and fine-tuning procedure for improving LLMs in generating proofs for the correctness of Rust code. SAFE consists of t... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "4;3;3;4",
"contribution_avg": 3.5,
"presentation": "2;3;2;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.680812"
} | {
"id": "tEI7grfbTy",
"metareview": "This paper considers learning how to prove the correctness of Rust programs, an important programming language for which there is very little training data, by bootstrapping both specifications and proofs. This problem is more broadly emblematic of the need to produce verified c... | {
"decision": "Accept (Poster)"
} |
2OMyAFjiJJ | 2405.20879v2 | Flow matching achieves almost minimax optimal convergence | {
"content": "## Abstract\n\nAbstract Flow matching (FM) has gained significant attention as a simulation-free generative model. Unlike diffusion models, which are based on stochastic differential equations, FM employs a simpler approach by solving an ordinary differential equation with an initial condition from a no... | [
{
"id": "w9avvZGAGP",
"initial_rating": 6,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "The paper provides estimates for the 2-Wasserstein distance for the sample-based distribution obtained in the Flow-Matching framework relative to the exact distri... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "5;3;4;3",
"confidence_avg": 3.75,
"soundness": "3;3;2;4",
"soundness_avg": 3,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.682085"
} | {
"id": "iGl9XuR3St",
"metareview": "The paper shows flow matching, or ODE-based generative models, achieves almost minimax optimal convergence, which complements the prior literature (e.g., Oko et al) on the minimax optimality of SDE-based generative models. Given the significance of flow matching, this result is ... | {
"decision": "Accept (Poster)"
} |
2Oh2EOcFSO | 2408.05284v2 | Can a Bayesian oracle prevent harm from an agent? | {
"content": "## Abstract\n\nAbstract Is there a way to design powerful AI systems based on machine learning methods that would satisfy probabilistic safety guarantees? With the long-term goal of obtaining a probabilistic guarantee that would apply in every context, we consider estimating a context-dependent bound on... | [
{
"id": "JvzstXgFIj",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper explores the problem of designing AI systems that satisfy probabilistic safety guarantees. Within a Bayesian framework and given the safety specificati... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "3;3;4;2",
"confidence_avg": 3,
"soundness": "4;3;3;3",
"soundness_avg": 3.25,
"contribution": "2;3;2;2",
"contribution_avg": 2.25,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.683195"
} | {
"id": "47YnKT9MpX",
"metareview": "This paper seeks to create practical safety mechanisms for AI by introducing a Bayesian framework.\nThe reviewers agreed that the topic is important, given the potential implications for AI safety (e.g., Reviewer bi5T noted that the framework tackles “a significant problem in AI... | {
"decision": "Reject"
} |
2P4p4RxUxT | 2410.03406v2 | Conformal confidence sets for biomedical image segmentation | {
"content": "## Abstract\n\nAbstract We develop confidence sets which provide spatial uncertainty guarantees for the output of a black-box machine learning model designed for image segmentation. To do so we adapt conformal inference to the imaging setting, obtaining thresholds on a calibration dataset based on the d... | [
{
"id": "qj8cbBdYeh",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "Authors develop confidence sets providing spatial uncertainty guarantees for outputs of a black-box machine learning model designed for image segmentation. Specif... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;3;2;4",
"confidence_avg": 3.25,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.683899"
} | {
"id": "Nl0aqYQMMy",
"metareview": "1. The method is only evaluated on a single dataset and focuses exclusively on binary segmentation. \n2. The manuscript lacks a thorough comparison with existing conformal prediction methods for image segmentation. \n3. Some aspects of the methodology, particularly how the algo... | {
"decision": "Reject"
} |
2R7498e2Tx | 2409.20296v1 | PersonalLLM: Tailoring LLMs to Individual Preferences | {
"content": "## Abstract\n\nAbstract As LLMs become capable of complex tasks, there is growing potential for personalized interactions tailored to the subtle and idiosyncratic preferences of the user.\nWe present a public benchmark, PersonalLLM , focusing on adapting LLMs to provide maximal benefits for a particular... | [
{
"id": "3CyRvdPiEQ",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper introduces PersonalLLM, a public benchmark designed to personalize Large Language Models (LLMs) to better align with individual user preferences. The be... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "5;4;4;3",
"confidence_avg": 4,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;4",
"contribution_avg": 2.5,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.684799"
} | {
"id": "h2UqVhSGeg",
"metareview": "This paper introduces PERSONALLLM, a novel dataset for advancing research in the personalization AI domain. The dataset captures user preferences through prompts accompanied by eight responses generated by various large language models (LLMs), including GPT-4 and Claude 3. To be... | {
"decision": "Accept (Poster)"
} |
2TuUXtLGhT | 2410.05690v1 | Long-Context Linear System Identification | {
"content": "## Abstract\n\nAbstract This paper addresses the problem of long-context linear system identification, where the state x t subscript 𝑥 𝑡 x_{t} italic_x start_POSTSUBSCRIPT italic_t end_POSTSUBSCRIPT of a dynamical system at time t 𝑡 t italic_t depends linearly on previous states x s subscript 𝑥 𝑠 x... | [
{
"id": "VOMLpWnnCy",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors study the problem of identifying long-context linear systems where the state at any given time depends on a sequence of previous states over an extend... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;2;3;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.685841"
} | {
"id": "Y0nafV1j1K",
"metareview": "This paper addresses the problem of linear system identification within a long-context framework. Specifically, it makes the following contributions:\n\n1. Theoretical Guarantee for the Constrained Least Squares Estimator: Under mild assumptions on the design matrix and sub-Gaus... | {
"decision": "Accept (Poster)"
} |
2U8owdruSQ | 2402.15163v3 | Has the Deep Neural Network learned the Stochastic Process? An Evaluation Viewpoint | {
"content": "## Abstract\n\nAbstract This paper presents the first systematic study of evalution of Deep Neural Network (DNN) designed and trained to predict the evolution of a stochastic dynamical system, using wildfire prediction as a case study. We show that traditional evaluation methods based on threshold based... | [
{
"id": "H9PaE0SeEj",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "the metric of expected calibration error is introduced and studied as a way to capture fidelity of a learned representation to an underlying stochastic process (r... | {
"rating": "5;5;5;8;8",
"rating_avg": 6.2,
"confidence": "2;2;3;5;4",
"confidence_avg": 3.2,
"soundness": "2;3;2;3;4",
"soundness_avg": 2.8,
"contribution": "2;2;2;1;4",
"contribution_avg": 2.2,
"presentation": "3;1;2;3;4",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.686898"
} | {
"id": "3Itw6iCNkG",
"metareview": "The reviewers unanimously recommend acceptance (8-6-6-8-6). The paper presents a significant contribution for the evaluation of neural networks designed to forecast the evolution of stochastic complex systems. Reviewers recognize the importance of the work and the quality of the... | {
"decision": "Accept (Poster)"
} |
2XBPdPIcFK | 2308.10248v5 | Steering Language Models with Activation Engineering | {
"content": "## Abstract\n\nAbstract Prompt engineering and finetuning aim to maximize language model performance on a given metric (like toxicity reduction). However, these methods do not fully elicit a model’s capabilities. To reduce this gap, we introduce activation engineering : the inference-time modification o... | [
{
"id": "vPosGYaBNO",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "In this paper, the authors introduce a paradigm of controlling model outputs/behavior which they term activation engineering. In activation engineering, a user co... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;1;3;3",
"soundness_avg": 2.25,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.687650"
} | {
"id": "6m3mbcpaDy",
"metareview": "The paper is motivated by elicitation overhang - prompt engineering may not be able to elicit all the information from a language model. They introduce ActAdd, a method that modifies the inner activations of an LLM during the forward passes to elicit text with a specific propert... | {
"decision": "Reject"
} |
2XdRkRHBT9 | 2402.13501v1 | AVOIDING BARREN PLATEAUS VIA GAUSSIAN MIXTURE MODEL | {
"content": "## Abstract\n\nAbstract Variational quantum algorithms is one of the most representative algorithms in quantum computing, which has a wide range of applications in quantum machine learning, quantum simulation and other related fields. However, they face challenges associated with the barren plateau phen... | [
{
"id": "jGaSA62NJH",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper considers the variational quantum algorithms and deal with the barren plateau phenomenon. The new parameter initialization strategy is proposed combing... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "3;3;4;3",
"confidence_avg": 3.25,
"soundness": "1;1;3;2",
"soundness_avg": 1.75,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "1;1;2;2",
"presentation_avg": 1.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.688665"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
2ZK8zyIt7o | 2410.11817v1 | Improving Long-Text Alignment for Text-to-Image Diffusion Models | {
"content": "## Abstract\n\nAbstract The rapid advancement of text-to-image (T2I) diffusion models has enabled them to generate unprecedented results from given texts. However, as text inputs become longer, existing encoding methods like CLIP face limitations, and aligning the generated images with long texts become... | [
{
"id": "kax49zb3Nf",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a novel approach to enhance the alignment between long text descriptions and generated images in text-to-image diffusion models, introducing s... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;4;2",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.689545"
} | {
"id": "6GD06cH6Hq",
"metareview": "This paper proposes a novel text encoder to deal with long conditioning texts in text-to-image diffusion models. Long texts are divided into segments and processed separately. The authors finetune a stable diffusion model based on this encoding, using a CLIP-based preference opt... | {
"decision": "Accept (Poster)"
} |
2ZTnALzLyX | 2202.00519v2 | MotifExplainer: a Motif-based Graph Neural Network Explainer | {
"content": "## Abstract\n\nAbstract We consider the explanation problem of Graph Neural Networks (GNNs). Most existing GNN explanation methods identify the most important edges or nodes but fail to consider substructures, which are more important for graph data. One method considering subgraphs tries to search all ... | [
{
"id": "eKyRpZFo9R",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a GNN explainer that uses motifs as the unit of explanation. By decomposing representations based on extracted motifs, it produces subgraph ex... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;1;2",
"soundness_avg": 1.75,
"contribution": "2;2;1;3",
"contribution_avg": 2,
"presentation": "1;3;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.690303"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
2bn7gayfz9 | 2406.04848v2 | CTBench: A Library and Benchmark for Certified Training | {
"content": "## Abstract\n\nAbstract Training certifiably robust neural networks is an important but challenging task. While many algorithms for (deterministic) certified training have been proposed, they are often evaluated on different training schedules, certification methods, and systematically under-tuned hyper... | [
{
"id": "PHqrCIdJeX",
"initial_rating": 5,
"confidence": 4,
"soundness": 4,
"contribution": 2,
"presentation": 4,
"summary": "The paper presents CTBENCH, a standardized library and benchmark designed to fairly evaluate certified training algorithms for neural networks, addressing the inc... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;4;4;3",
"soundness_avg": 3.25,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;4;4;4",
"presentation_avg": 3.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.690760"
} | {
"id": "xU06RGVYWM",
"metareview": "This paper develops a benchmark and library for several representative deterministic certified training algorithms for fair comparison on hyperparameters, training schedules and (exact) certification methods. The reviewers generally agree that this paper is well-written, however... | {
"decision": "Reject"
} |
2c7pfOqu9k | 2404.00242v3 | DeFT: Decoding with Flash Tree-attention for Efficient Tree-structured LLM Inference | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) are increasingly employed for complex tasks that process multiple generation calls in a tree structure with shared prefixes of tokens, including few-shot prompting, multi-step reasoning, speculative decoding, etc.\nHowever, existing inference systems ... | [
{
"id": "PwIWYjlw0p",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "Tree-structured decoding is gaining more popularity in LLM serving due to the presence of applications such as multi-step reasoning and speculative decoding. Exis... | {
"rating": "6;8;8;8",
"rating_avg": 7.5,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "3;4;3;3",
"soundness_avg": 3.25,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.691600"
} | {
"id": "Qb96kedKnG",
"metareview": "This paper presents DeFT, a novel algorithm for enhancing tree-based decoding in LLM inference. It targets the inefficiencies of existing systems, such as redundant KV cache access and poor load balancing. By introducing techniques like KV-Guided Grouping and Flattened Tree KV S... | {
"decision": "Accept (Spotlight)"
} |
2eFq6S35iB | 2408.04591v1 | HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts | {
"content": "## Abstract\n\nAbstract Generalized Category Discovery (GCD) is a challenging task in which, given a partially labelled dataset, models must categorize all unlabelled instances, regardless of whether they come from labelled categories or from new ones.\nIn this paper, we challenge a remaining assumption... | [
{
"id": "wj1zyCvgGp",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a new challenge for Generalized Category Discovery, which requires model to categorize unlabeled data in the presence of domain shifts. Trad... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "3;2;2;2",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.692620"
} | {
"id": "NhRE9twbgW",
"metareview": "The paper introduces HiLo, a novel framework for Generalized Category Discovery (GCD) under domain shifts. It disentangles semantic and domain features using mutual information minimization, enhances learning with PatchMix-based contrastive learning, and integrates curriculum le... | {
"decision": "Accept (Poster)"
} |
2ea5TNVR0c | 2404.02078v1 | Advancing LLM Reasoning Generalists with Preference Trees | {
"content": "## Abstract\n\nAbstract We introduce Eurus , a suite of large language models (LLMs) optimized for reasoning.\nFinetuned from Mistral-7B and CodeLlama-70B, Eurus models achieve state-of-the-art results among open-source models on a diverse set of benchmarks covering mathematics, code generation, and log... | [
{
"id": "DSaOq8Zmac",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors explore improving large language model reasoning through the curation of high quality training data for that reasoning.\nThis data (UltraInteract) con... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "3;3;2;4",
"contribution_avg": 3,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.693516"
} | {
"id": "ghwPNrfH25",
"metareview": "This paper presents a suite of large language models that achieve state-of-the-art performance in reasoning tasks. Key contributions include the introduction of a novel dataset featuring multi-turn preference trees designed for reasoning, and a reward modeling objective tailored... | {
"decision": "Accept (Poster)"
} |
2edigk8yoU | 2409.15647v2 | Looped Transformers for Length Generalization | {
"content": "## Abstract\n\nAbstract Recent work has shown that Transformers trained from scratch can successfully solve various arithmetic and algorithmic tasks, such as adding numbers and computing parity. While these Transformers generalize well on unseen inputs of the same length, they struggle with length gener... | [
{
"id": "zdcgZb1Jue",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "- This work studies the efficacy of Looped Transformers for Length Generalization of several algorithmic tasks whose computation complexity is known (as a functio... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "4;3;4;3",
"soundness_avg": 3.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.694358"
} | {
"id": "o61iqgOP0U",
"metareview": "This paper studies length generalization in Transformers through the lens of looped architectures, showing that Transformers with repeated layers can effectively generalize to longer sequences when trained appropriately. The reviewers appreciated the paper's clear writing, empir... | {
"decision": "Accept (Poster)"
} |
2efNHgYRvM | 2405.15325v2 | On the Identification of Temporal Causal Representation with Instantaneous Dependence | {
"content": "## Abstract\n\nAbstract Temporally causal representation learning aims to identify the latent causal process from time series observations, but most methods require the assumption that the latent causal processes do not have instantaneous relations. Although some recent methods achieve identifiability i... | [
{
"id": "82iJJTJC20",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes IDOL, a framework for achieving identifiability in sequential latent variable models with instantaneous dependencies. The authors establish id... | {
"rating": "6;6;8",
"rating_avg": 6.666666666666667,
"confidence": "3;4;2",
"confidence_avg": 3,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "3;2;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.695274"
} | {
"id": "uN8b3aVlpm",
"metareview": "The authors propose a framework to identify temporally causal relations with instantaneous dependencies. The three reviewers all voted to accept the paper, noting that the problem is important and motivated well and that the incorporation of instantaneous effects is a signific... | {
"decision": "Accept (Oral)"
} |
2fZ9iOVzpR | 2405.14021v2 | A Study of Posterior Stability for Time-Series Latent Diffusion | {
"content": "## Abstract\n\nAbstract Latent diffusion has demonstrated promising results in image generation and permits efficient sampling. However, this framework might suffer from the problem of posterior collapse when applied to time series. In this paper, we first show that posterior collapse will reduce latent... | [
{
"id": "GjpwRDQ5cC",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper addresses the problem of posterior collapse in latent diffusion models, specifically when applied to time series data. The authors provide a systematic... | {
"rating": "3;5;5;5;5;8",
"rating_avg": 5.166666666666667,
"confidence": "4;3;2;3;5;4",
"confidence_avg": 3.5,
"soundness": "2;2;3;2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;2;2;3;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;3;3;2;3",
"presentation_avg": 2... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.696110"
} | {
"id": "GCgux5hzwd",
"metareview": "This paper proposed a metric to measure the posterior collapse and introduced a notion of the posterior collapse based on it. They then define an enhanced framework which can alleviate the posterior collapse problem. I found that their posterior collapse is different from the mo... | {
"decision": "Reject"
} |
2fgzf8u5fP | 2408.08252v5 | Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-Based Decoding | {
"content": "## Abstract\n\nAbstract Diffusion models excel at capturing the natural design spaces of images, molecules, and biological sequences of DNA, RNA, and proteins. However, for many applications from biological research to biotherapeutic discovery, rather than merely generating designs that are natural, we ... | [
{
"id": "ddfPEfZDnE",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a method for diffusion models to sample data that is both within target distribution and maximizing some downstream reward function. The probl... | {
"rating": "3;3;3;5;5",
"rating_avg": 3.8,
"confidence": "4;5;4;3;2",
"confidence_avg": 3.6,
"soundness": "3;2;1;3;3",
"soundness_avg": 2.4,
"contribution": "2;1;2;3;2",
"contribution_avg": 2,
"presentation": "2;2;2;3;3",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.696985"
} | {
"id": "oX4kwd2waU",
"metareview": "This work presents a unified framework for guidance in diffusion models, encompassing both discrete and continuous settings, with minimal additional training. It extends applicability to domains where downstream rewards may not be differentiable. The proposed method, SVDD (MC an... | {
"decision": "Reject"
} |
2fojNANZSv | 2405.16156v1 | Mixture of In-Context Prompters for Tabular PFNs | {
"content": "## Abstract\n\nAbstract Recent benchmarks found In-Context Learning (ICL) outperforms both deep learning and tree-based algorithms on small tabular datasets. However, on larger datasets, ICL for tabular learning cannot run without severely compromising performance, due to its quadratic space and time co... | [
{
"id": "NKcIW6C5DI",
"initial_rating": 3,
"confidence": 2,
"soundness": 3,
"contribution": 1,
"presentation": 3,
"summary": "The paper proposes a mixture of experts approach for in-context learning on tabular data. Each expert in the mixture is a K-means cluster and the model routes the... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "2;3;2",
"confidence_avg": 2.3333333333333335,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "1;3;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.697895"
} | {
"id": "WW5lxpMmq0",
"metareview": "(a) Summary of Scientific Claims and Findings\n\nThe paper presents MixturePFN, an enhancement of Sparse Mixture of Experts tailored for tabular Prior-Fitted Networks (TabPFNs). MixturePFN leverages specialized In-Context Learning (ICL) experts, applied to clusters of tabular da... | {
"decision": "Accept (Poster)"
} |
2gTEW29qsM | 2410.07836v3 | Masked Generative Priors Improve World Models Sequence Modelling Capabilities | {
"content": "## Abstract\n\nAbstract Deep Reinforcement Learning (RL) has become the leading approach for creating artificial agents in complex environments. Model-based approaches, which are RL methods with world models that predict environment dynamics, are among the most promising directions for improving data ef... | [
{
"id": "86HJuzDCR4",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper, Masked Generative Priors Improve World Models Sequence Modelling Capabilities, introduces GIT-STORM, an extension of the STORM architecture, incorporat... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;2;2;2",
"soundness_avg": 2,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.698670"
} | {
"id": "pvem2GeP5x",
"metareview": "While GIT-STORM shows some promising results in improving world model sequence modeling through masked generative priors, several critical concerns remain insufficiently addressed. The authors' claim of being the first to apply transformer-based world models to continuous action... | {
"decision": "Reject"
} |
2h1siDrSMl | 2410.08876v2 | RoRA-VLM: Robust Retrieval-Augmented Vision Language Models | {
"content": "## Abstract\n\nAbstract Though vision-language models (VLMs) have demonstrated impressive capabilities as general-purpose visual assistants, they still exhibit inferior performance on knowledge-intensive tasks such as information-seeking visual question answering, primarily due to the challenge of accur... | [
{
"id": "F9WHqMfSxa",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces RORA-VLM, a retrieval-augmented framework designed to enhance Vision-Language Models (VLMs) by addressing two main challenges: managing multi... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;3;3",
"confidence_avg": 3.3333333333333335,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "1;2;3",
"contribution_avg": 2,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.708463"
} | {
"id": "I7qnRmKJhz",
"metareview": "The paper proposes a retrieval-augmented method for knowledge-intensive tasks to make more relevant use of visual information. The reviewers praise the extensive experiments. However they raise numerous concerns about the method and experiments; some raise a concern about novelt... | {
"decision": "Reject"
} |
2hKDQ20zDa | 2405.11597v1 | Language Reconstruction with Brain Predictive Coding from fMRI Data | {
"content": "## Abstract\n\nAbstract Many recent studies have shown that the perception of speech can be decoded from brain signals and subsequently reconstructed as continuous language.\nHowever, there is a lack of neurological basis for how the semantic information embedded within brain signals can be used more ef... | [
{
"id": "a5mkBp4lV1",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper describes a decoding method \"PredFT\" that uses a main decoding network and a side network to perform decoding from fMRI recordings of subjects listeni... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "5;4;5;4",
"confidence_avg": 4.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;2",
"contribution_avg": 2.25,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.709113"
} | {
"id": "ItLTrahT40",
"metareview": "This works aims to improve the SoTA of text decoding (language reconstruction) from fMRI with semantics.\n\nReviewers have raised a number of concerns concerning mainly the evaluation of the performance of the method arguing that, as is, the evidence of improved text decoding wi... | {
"decision": "Reject"
} |
2iPvFbjVc3 | 2402.17969v1 | Vision Language Model Based Caption Evaluation Method Leveraging Visual Context Extraction | {
"content": "## Abstract\n\nAbstract Given the accelerating progress of vision and language modeling, accurate evaluation of machine-generated image captions remains critical. In order to evaluate captions more closely to human preferences, metrics need to discriminate between captions of varying quality and content... | [
{
"id": "2foSGTRjgl",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents VisCE2, a vision-language model-based caption evaluation method designed to evaluate captions in a manner that aligns more closely with human p... | {
"rating": "3;3;3;3;5",
"rating_avg": 3.4,
"confidence": "4;4;4;4;2",
"confidence_avg": 3.6,
"soundness": "3;2;3;2;2",
"soundness_avg": 2.4,
"contribution": "2;2;2;2;2",
"contribution_avg": 2,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.709733"
} | {
"id": "Kp0GrCh0uI",
"metareview": "The paper introduces a vision-language model-based caption evaluation method. \nReviewers have raised major concerns on inaccurate literature review, limited novelty, insufficient comparisons, ablations, and analysis. All five reviewers recommended rejection. No rebuttal was pro... | {
"decision": "Reject"
} |
2iYVBqRHK4 | 2410.05527v1 | DOPL: Direct Online Preference Learning for Restless Bandits with Preference Feedback | {
"content": "## Abstract\n\nAbstract Restless multi-armed bandits (RMAB) has been widely used to model constrained sequential decision making problems, where the state of each restless arm evolves according to a Markov chain and each state transition generates a scalar reward. However, the success of RMAB crucially ... | [
{
"id": "XaP04lyllF",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This work studies a new problem set-up, PREF-RMAB.\nFor me, the problem set-up is very incremental. It is quite similar to duelling bandits. The proposed set-up i... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;3;4",
"contribution_avg": 3,
"presentation": "2;2;2;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.710702"
} | {
"id": "SV1EBLUjPw",
"metareview": "The paper introduces PREF-RMAB, a new variant of the Restless Multi-Armed Bandits (RMAB) problem that leverages preference feedback instead of direct reward estimations. This approach is particularly relevant for applications such as app marketing and CPAP treatment, where obtai... | {
"decision": "Accept (Poster)"
} |
2jTdHYuguF | 2409.02813v2 | MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmark | {
"content": "## Abstract\n\nAbstract This paper introduces MMMU-Pro, a robust version of the Massive Multi-discipline Multimodal Understanding and Reasoning (MMMU) benchmark. MMMU-Pro rigorously assesses multimodal models’ true understanding and reasoning capabilities through a three-step process based on MMMU: (1) ... | [
{
"id": "dLlqaSYo4l",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces MMMU-Pro, a more robust version of the MMMU benchmark. MMMU-Pro aims to more accurately assess multimodal models' true understanding and rea... | {
"rating": "5;5;5;6;6",
"rating_avg": 5.4,
"confidence": "5;3;4;3;4",
"confidence_avg": 3.8,
"soundness": "2;3;2;3;3",
"soundness_avg": 2.6,
"contribution": "2;3;2;3;3",
"contribution_avg": 2.6,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.711751"
} | {
"id": "S9qbWIpQD9",
"metareview": "This paper introduces MMMU-Pro, an enhanced multimodal benchmark that addresses limitations of the original MMMU by ensuring reliance on multimodal inputs, expanding answer options, and introducing vision-only settings, providing a more rigorous evaluation of multimodal models' ... | {
"decision": "Reject"
} |
2kje23LSOE | 2208.01958v1 | Moment Constrained Optimal Transport for Control Applications | {
"content": "## Abstract\n\nAbstract Optimal transport is now a standard tool for solving many problems in statistics and machine learning. The optimal “transport of probability measures” is also a recurring theme in stochastic control and distributed control, where in the latter application the probability measure ... | [
{
"id": "oFF34T8qfS",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper considers using optimal transport (OT) in mean field control, with constraints on the moments of the distributions. An algorithm (Sinkhorn algorithm) i... | {
"rating": "3;3;3;3;3;6",
"rating_avg": 3.5,
"confidence": "3;2;2;3;3;3",
"confidence_avg": 2.6666666666666665,
"soundness": "1;2;2;2;2;3",
"soundness_avg": 2,
"contribution": "2;2;2;2;2;3",
"contribution_avg": 2.1666666666666665,
"presentation": "1;1;1;1;2;3",
"presentation_avg": 1.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.712430"
} | {
"id": "djy6lsxWUm",
"metareview": "While the reviewers did not find any disqualifying technical errors for the submission, and are appreciative of its general contributions, its connections with machine learning remain tenuous, and the authors have not sufficiently endeavored to make the connection to problems in... | {
"decision": "Reject"
} |
2mGFmAQWUI | 2410.19811v1 | ControlAgent: Automating Control System Design via Novel Integration of LLM Agents and Domain Expertise | {
"content": "## Abstract\n\nAbstract Control system design is a crucial aspect of modern engineering with far-reaching applications across diverse sectors including aerospace, automotive systems, power grids, and robotics. Despite advances made by Large Language Models (LLMs) in various domains, their application in... | [
{
"id": "I6nIIkkcII",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 4,
"summary": "This paper introduces ControlAgent, a framework that automates control system design by integrating large language model (LLM) agents with domain expertise. The f... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;3;4",
"confidence_avg": 3.3333333333333335,
"soundness": "1;2;2",
"soundness_avg": 1.6666666666666667,
"contribution": "1;2;3",
"contribution_avg": 2,
"presentation": "2;2;4",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.713301"
} | {
"id": "ZZGskUlVSD",
"metareview": "This paper's core strength is its demonstration that LLMs can be used to design controllers for a wide variety of different systems with specific stability, phase margin, and settling times. The system, ControlAgent, runs in a loop where the LLM designed controller is run and th... | {
"decision": "Reject"
} |
2mqb8bPHeb | 2402.14167v1 | T-Stitch: Accelerating Sampling in Pre-Trained Diffusion Models with Trajectory Stitching | {
"content": "## Abstract\n\nAbstract Sampling from diffusion probabilistic models (DPMs) is often expensive for high-quality image generation and typically requires many steps with a large model.\nIn this paper, we introduce sampling Trajectory Stitching ( T-Stitch ), a simple yet efficient technique to improve the ... | [
{
"id": "3ifioWY7zT",
"initial_rating": 8,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces T-Stitch, a training-free approach to accelerate sampling in diffusion models by strategically utilizing different-sized models across the d... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "3;4;3;5",
"confidence_avg": 3.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"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": "2026-01-14T22:16:00.714154"
} | {
"id": "ZGeRN9dlt6",
"metareview": "This paper introduces a novel and training-free approach to accelerating the sampling process of diffusion models by leveraging small diffusion models. The proposed method is both simple and effective, demonstrating effectiveness across various tasks, including large-scale text-... | {
"decision": "Accept (Poster)"
} |
2ofVtMvRil | 2410.01022v2 | Learning grid cells by predictive coding | {
"content": "## Abstract\n\nAbstract Grid cells in the medial entorhinal cortex (MEC) of the mammalian brain exhibit a strikingly regular hexagonal firing field over space. These cells are learned after birth and are thought to support spatial navigation but also more abstract computations. Although various computat... | [
{
"id": "pQ2F0Tq7Cn",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 1,
"presentation": 3,
"summary": "The authors investigate how a temporally-dependent version of predictive coding can extract compact latent spaces in the form of periodic grid activity from tempo... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "3;3;4;4",
"confidence_avg": 3.5,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "1;2;1;3",
"contribution_avg": 1.75,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.714997"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
2pJpFtdVNe | 2406.18450v1 | Preference Elicitation for Offline Reinforcement Learning | {
"content": "## Abstract\n\nAbstract Applying reinforcement learning (RL) to real-world problems is often made challenging by the inability to interact with the environment and the difficulty of designing reward functions. Offline RL addresses the first challenge by considering access to an offline dataset of enviro... | [
{
"id": "Fmtsa0bQ9G",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper studies preference-based reinforcement learning (PbRL) in offline setting, in which the agent utilizes a fixed trajectory dataset for policy learning a... | {
"rating": "5;6;6;6;8",
"rating_avg": 6.2,
"confidence": "3;4;3;2;2",
"confidence_avg": 2.8,
"soundness": "3;3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;2;2",
"contribution_avg": 2.6,
"presentation": "4;3;3;3;3",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.715868"
} | {
"id": "3OoZQCSSJh",
"metareview": "This paper studies preference-based reinforcement learning (PbRL) in offline setting, a relatively unexplored area.\nThe authors show the sample complexity effectiveness of their approach.\nThe main contribution of the paper is theoretical; however, they derive a practical algor... | {
"decision": "Accept (Poster)"
} |
2pNLknCTvG | 2410.03284v1 | uniINF: Best-of-Both-Worlds Algorithm for Parameter-Free Heavy-Tailed MABs | {
"content": "## Abstract\n\nAbstract In this paper, we present a novel algorithm, uniINF , for the Heavy-Tailed Multi-Armed Bandits (HTMAB) problem, demonstrating robustness and adaptability in both stochastic and adversarial environments. Unlike the stochastic MAB setting where loss distributions are stationary wit... | [
{
"id": "s0zXddi1KN",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper studies Heavy-Tailed MultiArmed Bandits (HTMAB) problem. The main contribution of the paper is to design an optimal algorithm that achieves both Best of... | {
"rating": "6;6;6;6",
"rating_avg": 6,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.716907"
} | {
"id": "CPHGo0mNqx",
"metareview": "This paper proposes a best-of-both-worlds algorithm for the heavy-tailed multi-armed bandit problem, which achieves optimal performance in both stochastic and adversarial environments. A significant strength of the proposed algorithm is its adaptability to the (unknown) heavy-ta... | {
"decision": "Accept (Spotlight)"
} |
2prShxdLkX | 2406.00434v2 | MoDGS: Dynamic Gaussian Splatting from Casually-captured Monocular Videos | {
"content": "## Abstract\n\nAbstract In this paper, we propose MoDGS, a new pipeline to render novel views of dynamic scenes from a casually captured monocular video. Previous monocular dynamic NeRF or Gaussian Splatting methods strongly rely on the rapid movement of input cameras to construct multiview consistency ... | [
{
"id": "jEYfHT7GvX",
"initial_rating": 8,
"confidence": 5,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "The paper introduces MoDGS (Monocular Dynamic Gaussian Splatting), a novel approach for rendering dynamic 3D scenes from casually captured monocular videos, overc... | {
"rating": "3;5;8;8",
"rating_avg": 6,
"confidence": "4;4;3;5",
"confidence_avg": 4,
"soundness": "2;2;4;4",
"soundness_avg": 3,
"contribution": "2;2;4;4",
"contribution_avg": 3,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.717906"
} | {
"id": "SHCW9JOGm1",
"metareview": "This paper presents MoDGS, a dynamic scene modeling algorithm using casually captured monocular videos. To overcome the rapid camera motion assumption in previous methods, it integrates single-view depth estimation, 3D-aware initialization, and robust ordinal depth loss, and it ... | {
"decision": "Accept (Poster)"
} |
2pvMZKGYDR | 2408.03092v1 | Extend Model Merging from Fine-Tuned to Pre-Trained Large Language Models via Weight Disentanglement | {
"content": "## Abstract\n\nAbstract Merging Large Language Models (LLMs) aims to amalgamate multiple homologous LLMs into one with all the capabilities. Ideally, any LLMs sharing the same backbone should be mergeable, irrespective of whether they are Fine-Tuned (FT) with minor parameter changes or Pre-Trained (PT) ... | [
{
"id": "1arFJJLqMh",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "Merging multiple LLMs, particularly those with substantial parameter shifts from pre-training (PT), presents challenges for traditional merging methods. To addres... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "3;4;2",
"confidence_avg": 3,
"soundness": "3;2;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"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": "2026-01-14T22:16:00.718571"
} | {
"id": "YWiuI9wC8R",
"metareview": "This paper tackles the model merging problem between PT and FT LLMs. The approach focuses on merging homogeneous models that are of a significant parameter divergence. The basic idea is to separate the weight into the magnitude and the direction components to automatically deter... | {
"decision": "Reject"
} |
2rWbKbmOuM | 2410.10563v2 | MEGA-Bench: Scaling Multimodal Evaluation to over 500 Real-World Tasks | {
"content": "## Abstract\n\nAbstract We present MEGA-Bench , an evaluation suite that scales multimodal evaluation to over 500 real-world tasks, to address the highly heterogeneous daily use cases of end users.\nOur objective is to optimize for a set of high-quality data samples that cover a highly diverse and rich ... | [
{
"id": "SnltjtjctU",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work presents a new benchmark for multimodal LLMs. The authors attempt to create a novel, diverse, comprehensive benchmark for vision-language reasoning usin... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "4;3;4;3",
"confidence_avg": 3.5,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "3;2;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": "2026-01-14T22:16:00.719972"
} | {
"id": "8gH9wg8G2S",
"metareview": "The paper introduces MEGA-BENCH, a comprehensive multimodal benchmark designed to evaluate the diverse capabilities of vision-language models across over 500 tasks. The major advantages of the benchmark are as follows: (i) scope and scale: it includes 507 realistic tasks with ov... | {
"decision": "Accept (Poster)"
} |
2tIyA5cri8 | 2410.01280v1 | Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models | {
"content": "## Abstract\n\nAbstract In-context learning, the ability to adapt based on a few examples in the input prompt, is a ubiquitous feature of large language models (LLMs). However, as LLMs’ in-context learning abilities continue to improve, understanding this phenomenon mechanistically becomes increasingly ... | [
{
"id": "XfHLBIt8ZA",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "The paper investigates whether Llama 3 70B has internal representations that support temporal difference learning. First, it demonstrates that Llama can solve RL ... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;1;3;4",
"confidence_avg": 3,
"soundness": "1;2;3;4",
"soundness_avg": 2.5,
"contribution": "3;2;3;4",
"contribution_avg": 3,
"presentation": "4;2;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.721292"
} | {
"id": "bKYZYMSpPZ",
"metareview": "The paper investigates whether Llama 3 70B has internal representations that support temporal difference learning. First, it demonstrates that Llama can solve RL tasks significantly better than chance. Next, it trains a sparse autoencoder (SAE) and finds features correlated with... | {
"decision": "Accept (Poster)"
} |
2vMGPrk0SW | 2406.07163v1 | FaceGPT: Self-supervised Learning to Chat about 3D Human Faces | {
"content": "## Abstract\n\nAbstract We introduce FaceGPT, a self-supervised learning framework for Large Vision-Language Models (VLMs) to reason about 3D human faces from images and text. Typical 3D face reconstruction methods are specialized algorithms that lack semantic reasoning capabilities. FaceGPT overcomes t... | [
{
"id": "bxiveN7DmN",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper describes a method where a VLM is trained with LORA to be adapted for the task of 3D face reconstruction. The VLM is supposed to provide textual inform... | {
"rating": "3;3;5",
"rating_avg": 3.6666666666666665,
"confidence": "4;4;5",
"confidence_avg": 4.333333333333333,
"soundness": "2;2;2",
"soundness_avg": 2,
"contribution": "2;2;2",
"contribution_avg": 2,
"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": "2026-01-14T22:16:00.722008"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.