paper_id string | arxiv_id string | title string | markdown dict | reviews list | scores dict | metadata dict | meta_review dict | decision dict |
|---|---|---|---|---|---|---|---|---|
wQkERVYqui | 2411.02957v1 | Embedding Safety into RL: A New Take on Trust Region Methods | {
"content": "## Abstract\n\nAbstract Reinforcement Learning (RL) agents are able to solve a wide variety of tasks but are prone to producing unsafe behaviors.\nConstrained Markov Decision Processes (CMDPs) provide a popular framework for incorporating safety constraints.\nHowever, common solution methods often compr... | [
{
"id": "e9E9hKb8M6",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper presents a novel method for safe RL, i.e. solving CMDPs while ensuring safety during training. The method is based on modifying trust region methods (i.... | {
"rating": "3;3;5;6;8",
"rating_avg": 5,
"confidence": "4;4;2;3;3",
"confidence_avg": 3.2,
"soundness": "3;2;2;3;3",
"soundness_avg": 2.6,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.085651"
} | {
"id": "VV4xsXAQXQ",
"metareview": "Embedding Safety into RL: A New Take on Trust Region Methods\n\nSummary: The paper introduces a new approach for safe reinforcement learning (RL) using constrained trust region policy optimization (C-TRPO). The method modifies traditional trust region policy optimization (TRPO) ... | {
"decision": "Reject"
} |
wRbSdbGyfj | 2405.16672v1 | Transfer Learning Under High-Dimensional Graph Convolutional Regression Model for Node Classification | {
"content": "## Abstract\n\nAbstract Node classification is a fundamental task, but obtaining node classification labels can be challenging and expensive in many real-world scenarios. Transfer learning has emerged as a promising solution to address this challenge by leveraging knowledge from source domains to enhanc... | [
{
"id": "ImBjBGxbcl",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper addresses the challenge of node classification in high-dimensional settings, particularly in scenarios with limited labeled data. It introduces a novel... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;5;3;3",
"confidence_avg": 4,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.086700"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
wSErgkwDZO | 2410.13854v1 | Can MLLMs Understand the Deep Implication Behind Chinese Images? | {
"content": "## Abstract\n\nAbstract As the capabilities of Multimodal Large Language Models (MLLMs) continue to improve, the need for higher-order capability evaluation of MLLMs is increasing. However, there is a lack of work evaluating MLLM for higher-order perception and understanding of Chinese visual content.\n... | [
{
"id": "bhqcnUKWOZ",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a new benchmark, CII-Bench, for evaluating MLLMs on understanding Chinese image implications, which is an important capacity for MLLMs in ac... | {
"rating": "3;3;3;5;6",
"rating_avg": 4,
"confidence": "5;5;4;5;2",
"confidence_avg": 4.2,
"soundness": "3;3;2;3;3",
"soundness_avg": 2.8,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "3;3;3;4;3",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.087710"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
wUbum0nd9N | 2410.10414v1 | On Calibration of LLM-based Guard Models for Reliable Content Moderation | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) pose significant risks due to the potential for generating harmful content or users attempting to evade guardrails. Existing studies have developed LLM-based guard models designed to moderate the input and output of threat LLMs, ensuring adherence to ... | [
{
"id": "gupzR2PFcQ",
"initial_rating": 6,
"confidence": 2,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This work examines how calibration can affect and potentially improve LLM-based guard models. The study finds most guard models are poorly calibrated, especially ... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "4;4;2",
"confidence_avg": 3.3333333333333335,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.088676"
} | {
"id": "WVC4vhOz6N",
"metareview": "This is a solid empirical paper measuring present-day SotA LLM-based guardrails' (that classify prompts into e.g. jailbreak attempt vs not, and classify responses into e.g. toxic content or not) uncertainty quantification and calibration. All reviewers appreciated just how comp... | {
"decision": "Accept (Poster)"
} |
wUtCieKuQU | 2406.09179v1 | Towards Effective Evaluations and Comparison for LLM Unlearning Methods | {
"content": "## Abstract\n\nAbstract The compelling goal of eradicating undesirable data behaviors, while preserving usual model functioning, underscores the significance of machine unlearning within the domain of large language models (LLMs). Recent research has begun to approach LLM unlearning via gradient ascent ... | [
{
"id": "VhYriLXED1",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper tested 4 popular LLM unlearning methods on the TOFU benchmark dataset and observed that there are various tradeoffs (e.g. GA unlearns better while NPO r... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;2;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.089529"
} | {
"id": "hzivRj7PMn",
"metareview": "This work provides an evaluation framework for assessing LLM unlearning. The reviewers agree that the paper has made interesting conclusions and contributions, although the technical novelties and organizations are limited. It would be important for the authors to improve the pa... | {
"decision": "Accept (Poster)"
} |
wUtXB43Chi | 2410.01359v1 | FlashMask: Efficient and Rich Mask Extension of FlashAttention | {
"content": "## Abstract\n\nAbstract The computational and memory demands of vanilla attention scale quadratically with the sequence length N 𝑁 N italic_N , posing significant challenges for processing long sequences in Transformer models. FlashAttention alleviates these challenges by eliminating the 𝒪 ( N 2 ) �... | [
{
"id": "tYwrirqQWm",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "The paper introduces a novel compression scheme for the attention mask where only the boundary indices of the masks are stored for every column. For a specific se... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "4;5;4;4",
"confidence_avg": 4.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;2;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": "2026-01-14T22:16:04.090288"
} | {
"id": "OJlqMaIXkj",
"metareview": "The paper presents further optimization over Flash-Attention 2 for better memory management of the Attention mechanism, I think the novelty of the approach in this paper is limited. But given the reviewers are enthusiastic about it, it would be ok to accept.",
"additional_comm... | {
"decision": "Accept (Poster)"
} |
wWPiAjbR7a | 2410.06845v1 | MentalArena: Self-play Training of Language Models for Diagnosis and Treatment of Mental Health Disorders | {
"content": "## Abstract\n\nAbstract Mental health disorders are one of the most serious diseases in the world.\nMost people with such a disease lack access to adequate care, which highlights the importance of training models for the diagnosis and treatment of mental health disorders. However, in the mental health d... | [
{
"id": "Y1LmBfCMaX",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 4,
"summary": "The paper introduces MentalArena, a self-play framework designed to train language models for diagnosing and treating mental health disorders. Due to privacy conc... | {
"rating": "3;3;3;5;6",
"rating_avg": 4,
"confidence": "4;2;4;3;4",
"confidence_avg": 3.4,
"soundness": "2;2;2;3;2",
"soundness_avg": 2.2,
"contribution": "2;2;2;3;3",
"contribution_avg": 2.4,
"presentation": "2;3;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.091103"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
wWcNhS4g1U | 2410.16770v1 | The Scene Language: Representing Scenes with Programs, Words, and Embeddings | {
"content": "## Abstract\n\nAbstract We introduce the Scene Language, a visual scene representation that concisely and precisely describes the structure, semantics, and identity of visual scenes.\nIt represents a scene with three key components: a program that specifies the hierarchical and relational structure of e... | [
{
"id": "bjGr7ALkrt",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper utilizes the formal structure of the LISP language to define scenarios, allowing for precise expression of scenes and benefiting tasks such as generati... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.091907"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
wWnsoLhHwt | 2410.02064v1 | Inspection and Control of Self-Generated-Text Recognition Ability in Llama3-8b-Instruct | {
"content": "## Abstract\n\nAbstract It has been reported that LLMs can recognize their own writing. As this has potential implications for AI safety, yet is relatively understudied, we investigate the phenomenon, seeking to establish whether it robustly occurs at the behavioral level, how the observed behavior is a... | [
{
"id": "7vXCIa3DRv",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper investigates the self-recognition ability of large language models (LLMs), focusing on the Llama3-8b-Instruct model. The authors explore whether the mod... | {
"rating": "3;5;8;8",
"rating_avg": 6,
"confidence": "4;2;3;3",
"confidence_avg": 3,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.092564"
} | {
"id": "ZpIyJlqwNy",
"metareview": "This paper studies the ability of LLMs (specifically Llama3-8b-Instruct) to recognize text as being self generated. The authors found that LLaMA-3-8b-instruct (but not the base model) can distinguish texts created by it from texts created by humans, but not from texts created by... | {
"decision": "Accept (Poster)"
} |
wYJII5BRYU | 2310.13391v3 | Learning Successor Features with Distributed Hebbian Temporal Memory | {
"content": "## Abstract\n\nAbstract This paper presents a novel approach to address the challenge of online temporal memory learning for decision-making under uncertainty in non-stationary, partially observable environments. The proposed algorithm, Distributed Hebbian Temporal Memory (DHTM), is based on factor grap... | [
{
"id": "HRqf5ChIQY",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 2,
"summary": "This paper proposes a novel algorithm for learning successor features for reinforcement learning agents in non-stationary partially observable environments. The m... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "2;2;2;4",
"confidence_avg": 2.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;4",
"contribution_avg": 3,
"presentation": "1;2;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.093300"
} | {
"id": "lC004dMg67",
"metareview": "This paper introduces a novel, biologically-inspired algorithm (DHTM) designed for online sequence learning for decision making in non-stationary, partially observable environments. Leveraging factor graph formalism and Hebbian-like local learning rules, DHTM captures Successor ... | {
"decision": "Accept (Poster)"
} |
wYZ8rxwvMm | 2401.00330v3 | Two-Step Offline Preference-Based Reinforcement Learning with Constrained Actions | {
"content": "## Abstract\n\nAbstract Preference-based reinforcement learning (PBRL) in the offline setting has succeeded greatly in industrial applications such as chatbots. A two-step learning framework where one applies a reinforcement learning step after a reward modeling step has been widely adopted for the prob... | [
{
"id": "1DNTBg49I5",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The author discusses the success of preference-based reinforcement learning (PBRL) in offline settings, particularly in industrial applications like chatbots. A c... | {
"rating": "3;3;3;3",
"rating_avg": 3,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;1;2;2",
"soundness_avg": 1.75,
"contribution": "1;1;2;2",
"contribution_avg": 1.5,
"presentation": "1;2;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.094096"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
waHmD2i1dv | 2409.19306v1 | CausalVE: Face Video Privacy Encryption via Causal Video Prediction | {
"content": "## Abstract\n\nAbstract Advanced facial recognition technologies and recommender systems with inadequate privacy technologies and policies for facial interactions increase concerns about bioprivacy violations. With the proliferation of video and live-streaming websites, public-face video distribution an... | [
{
"id": "n92vWT2ykc",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper proposes CausalVE, a face video privacy protection framework that combines (1) a diffusion model for face swapping with facial guidance, (2) a video pre... | {
"rating": "3;3;5;5;6",
"rating_avg": 4.4,
"confidence": "4;4;2;5;4",
"confidence_avg": 3.8,
"soundness": "2;2;3;2;3",
"soundness_avg": 2.4,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "3;1;3;1;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.094853"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
waf6HreC53 | 2401.11576v3 | Quantum Architecture Search With Unsupervised Representation Learning | {
"content": "## Abstract\n\nAbstract Unsupervised representation learning presents new opportunities for advancing Quantum Architecture Search (QAS) on Noisy Intermediate-Scale Quantum (NISQ) devices. QAS is designed to optimize quantum circuits for Variational Quantum Algorithms (VQAs). Most QAS algorithms tightly ... | [
{
"id": "gwfSsnIqwC",
"initial_rating": 3,
"confidence": 5,
"soundness": 1,
"contribution": 1,
"presentation": 2,
"summary": "The authors focused on automatically designing the quantum circuit to reach acceptable accuracy while keeping the circuit at a very low circuit depth. They tried ... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;5;5;3",
"confidence_avg": 4.25,
"soundness": "2;2;1;2",
"soundness_avg": 1.75,
"contribution": "2;2;1;2",
"contribution_avg": 1.75,
"presentation": "2;3;2;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.095733"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
weM4YBicIP | 2409.02634v2 | Loopy: Taming Audio-Driven Portrait Avatar with Long-Term Motion Dependency | {
"content": "## Abstract\n\nAbstract With the introduction of diffusion-based video generation techniques, audio-conditioned human video generation has recently achieved significant breakthroughs in both the naturalness of motion and the synthesis of portrait details. Due to the limited control of audio signals in d... | [
{
"id": "iSH9uKNHne",
"initial_rating": 8,
"confidence": 5,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "The paper proposes an audio-only conditioned video diffusion model. The model consists of three key components: an inter- and intra-clip temporal module, and an a... | {
"rating": "6;8;8;8",
"rating_avg": 7.5,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "4;4;4;4",
"soundness_avg": 4,
"contribution": "3;3;4;4",
"contribution_avg": 3.5,
"presentation": "4;3;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.096424"
} | {
"id": "2o5ZCoD6ME",
"metareview": "The paper presents Loopy, an audio-driven portrait animation pipeline that demonstrates natural talking-head motion. The key innovation in Loopy against prior works is the use of diffusion-based synthesis using long-range motion frames that leads up to the currently synthesized ... | {
"decision": "Accept (Oral)"
} |
wfLuiDjQ0u | 2409.15700v1 | Making Text Embedders Few-Shot Learners | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) with decoder-only architectures demonstrate remarkable in-context learning (ICL) capabilities. This feature enables them to effectively handle both familiar and novel tasks by utilizing examples provided within their input context. Recognizing the pot... | [
{
"id": "4p0ZL5Todm",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes to train instruction-conditional embedding models to take few-shot examples as inputs. Previous work trained embedding models to take instructi... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "4;3;3;3",
"soundness_avg": 3.25,
"contribution": "4;3;2;3",
"contribution_avg": 3,
"presentation": "4;3;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.097125"
} | {
"id": "PF9R0ssAzR",
"metareview": "The reviewers agree that this paper makes a valuable contribution to text embedding research by introducing in-context learning capabilities to embedding models. The proposed approach, while conceptually straightforward, effectively leverages LLMs' ICL capabilities and demonstra... | {
"decision": "Accept (Poster)"
} |
wg1PCg3CUP | 2411.04330v1 | Scaling Laws for Precision | {
"content": "## Abstract\n\nAbstract Low precision training and inference affect both the quality and cost of language models, but current scaling laws do not account for this. In this work, we devise “precision-aware” scaling laws for both training and inference. We propose that training in lower precision reduces ... | [
{
"id": "Et5TAvrgTm",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This manuscript provides a thorough investigation into the impact of bit precision on inference performance and introduces a scaling law that correlates performan... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "3;3;4;3",
"confidence_avg": 3.25,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.097945"
} | {
"id": "6q6ad5rStb",
"metareview": "This paper investigates the scaling laws for precision, focusing on factors such as the number of parameters, tokens, pretraining precision, and inference precision. It provides a comprehensive background of an overview of quantization, a review of existing scaling laws for para... | {
"decision": "Accept (Oral)"
} |
wg3rBImn3O | 2410.01917v1 | Provably Accurate Shapley Value Estimation via Leverage Score Sampling | {
"content": "## Abstract\n\nAbstract Originally introduced in game theory, Shapley values have emerged as a central tool in explainable machine learning, where they are used to attribute model predictions to specific input features. However, computing Shapley values exactly is expensive: for a general model with n �... | [
{
"id": "FDqeKYLiz8",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This paper provides a new algorithm for approximating Shapley values with provable guarantees. Shapley values have widespread application in ML as a way of formal... | {
"rating": "5;8;8",
"rating_avg": 7,
"confidence": "2;3;4",
"confidence_avg": 3,
"soundness": "3;4;4",
"soundness_avg": 3.6666666666666665,
"contribution": "2;4;3",
"contribution_avg": 3,
"presentation": "2;4;4",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.099134"
} | {
"id": "xyZGUl7uKS",
"metareview": "This work considers the important Shapley value estimation problem. Authors show that for specific design matrices, there is a closed-form solution that can be used for leverage score sampling, which reduces the computational cost by orders of magnitude. All reviewers agree that... | {
"decision": "Accept (Spotlight)"
} |
wgDB1QuxIA | 2405.19440v4 | MGDA Converges under Generalized Smoothness, Provably | {
"content": "## Abstract\n\nAbstract Multi-objective optimization (MOO) is receiving more attention in various fields such as multi-task learning. Recent works provide some effective algorithms with theoretical analysis but they are limited by the standard L 𝐿 L italic_L -smooth or bounded-gradient assumptions, whi... | [
{
"id": "1v9BtqICYy",
"initial_rating": 8,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors analyze the problem of multi objective optimization (MOO), \n$$\\begin{align*}\nF^\\star = \\min_{x\\in \\mathbb{R}^d} F(x) = (f_1(x), f_2(x), \\ldots... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "3;2;2;2",
"confidence_avg": 2.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"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": "2026-01-14T22:16:04.100297"
} | {
"id": "IUfdadB3w0",
"metareview": "This paper analyzes the multi-objective optimization problem where the goal is to minimize the minimum of K different objectives. The standard measure for optimality in this case is called \\epsilon-conflict-avoidant (CA). The main contribution of the paper is to extend existing... | {
"decision": "Accept (Poster)"
} |
wgKW4U7ktq | 2409.13730v1 | VisScience: An Extensive Benchmark for Evaluating K12 Educational Multi-modal Scientific Reasoning | {
"content": "## Abstract\n\nAbstract Multi-modal large language models (MLLMs) have demonstrated promising capabilities across various tasks by integrating textual and visual information to achieve visual understanding in complex scenarios. Despite the availability of several benchmarks aims to evaluating MLLMs in t... | [
{
"id": "mEmjb79xoR",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces the VisScience benchmark, which is designed to evaluate the performance of multimodal large language models (MLLMs) in scientific reasoning t... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "2;3;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": "2026-01-14T22:16:04.101285"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
wgRQ2WAORJ | 2411.08923v1 | Aligning Visual Contrastive learning models via Preference Optimization | {
"content": "## Abstract\n\nAbstract Contrastive learning models have demonstrated impressive abilities to capture semantic similarities by aligning representations in the embedding space. However, their performance can be limited by the quality of the training data and its inherent biases. While Reinforcement Learn... | [
{
"id": "JPdl09gxSf",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 4,
"summary": "This paper introduces an alignment method designed for contrastive models, such as CLIP, using aligned and unaligned image-text pairs. In this setup, each image i... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "4;4;3;2",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;2;4",
"contribution_avg": 2.75,
"presentation": "1;2;4;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.102162"
} | {
"id": "xAaZAaPeMv",
"metareview": "The reviewers largely appreciated the paper's novel approach of applying Preference Optimization (PO) to enhance the CLIP contrastive learning model's robustness and reduce biases. Both Direct Preference Optimization (DPO) and Identity Preference Optimization (IPO) show promise ... | {
"decision": "Accept (Poster)"
} |
wh6pilyz2L | 2401.16845v3 | Chronicling Germany: An Annotated Historical Newspaper Dataset | {
"content": "## Abstract\n\nAbstract The correct detection of dense article layout and the recognition of characters in historical newspaper pages remains a challenging requirement for Natural Language Processing (NLP) and machine learning applications on historical newspapers in the field of digital history. Digita... | [
{
"id": "BsyFW2fuLM",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents the Chronicling Germany dataset. It consists in an annotated dataset of historical newspapers in German language. The dataset has been constru... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "5;5;3;4",
"confidence_avg": 4.25,
"soundness": "1;3;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"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": "2026-01-14T22:16:04.103037"
} | {
"id": "k0qWHdhdpI",
"metareview": "This paper discusses the challenges of detecting layouts and recognizing characters in historical German newspapers for in terms of Natural Language Processing and Machine Learning. The authors propose and discuss the *Chronicling Germany Dataset*, which consists of 693 annotate... | {
"decision": "Reject"
} |
whaO3482bs | 2410.09870v1 | ChroKnowledge: Unveiling Chronological Knowledge of Language Models in Multiple Domains | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have brought significant changes to many aspects of our lives.\nHowever, assessing and ensuring their chronological knowledge remains challenging.\nExisting approaches fall short in addressing the accumulative nature of knowledge, often relying on a s... | [
{
"id": "I0uaK7JS3c",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces CHROKNOWBENCH, a benchmark used to evaluate chronological knowledge in LLMs by distinguishing between evolving and static information. The a... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "1;4;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.103835"
} | {
"id": "y5tKruzIM8",
"metareview": "This paper introduces a benchmark dataset ChroKnowBench to evaluate chronologically accumulated knowledge of LLMs, and presents a prompting method to elicit chronological knowledge. The problem studies in this paper is important and interesting. The benchmark constructed in this... | {
"decision": "Accept (Poster)"
} |
wjgNVsbT3T | 2410.05262v1 | TurtleBench: Evaluating Top Language Models via Real-World Yes/No Puzzles | {
"content": "## Abstract\n\nAbstract As the application of Large Language Models (LLMs) expands, the demand for reliable evaluations increases. Existing LLM evaluation benchmarks primarily rely on static datasets, making it challenging to assess model performance in dynamic interactions with users. Moreover, these b... | [
{
"id": "lP4ejApZ2Y",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces TurtleBench, a novel evaluation benchmark designed to assess the reasoning capabilities of LLMs using data collected from an online game call... | {
"rating": "1;3;5;5;5",
"rating_avg": 3.8,
"confidence": "4;4;3;4;4",
"confidence_avg": 3.8,
"soundness": "1;2;3;2;2",
"soundness_avg": 2,
"contribution": "1;2;2;2;2",
"contribution_avg": 1.8,
"presentation": "2;3;3;3;2",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.104554"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
wkHcXDv7cv | 2410.02035v1 | Tuning Frequency Bias of State Space Models | {
"content": "## Abstract\n\nAbstract State space models (SSMs) leverage linear, time-invariant (LTI) systems to effectively learn sequences with long-range dependencies.\nBy analyzing the transfer functions of LTI systems, we find that SSMs exhibit an implicit bias toward capturing low-frequency components more effe... | [
{
"id": "MnmiPkfHLK",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This paper studies the frequency bias for State-space-models (SSMs). With strategies to select better initialization and Sobolev filtering (during training), this... | {
"rating": "6;8;8;8",
"rating_avg": 7.5,
"confidence": "3;2;4;3",
"confidence_avg": 3,
"soundness": "3;4;3;4",
"soundness_avg": 3.5,
"contribution": "3;3;3;4",
"contribution_avg": 3.25,
"presentation": "3;3;4;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.105296"
} | {
"id": "mB4RH5rIwB",
"metareview": "The authors are concerned with the observation that state space models contain an implicit bias for ignoring high-frequency components. After conducting a theoretical analysis to shed light into this phenomenon, they formulate a theory (related to initialization) which, in turn,... | {
"decision": "Accept (Spotlight)"
} |
wkbx7BRAsM | 2407.07356v1 | Autoregressive Transformers are Zero-Shot Video Imitators | {
"content": "## Abstract\n\nAbstract In-context learning for vision data has been underexplored compared with that in natural language. Previous works studied image in-context learning, urging models to generate a single image guided by demonstrations. In this paper, we propose and study video in-context learning, w... | [
{
"id": "s1iJuN5LH0",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a video generation based on autoregressive transformers trained with the objective of imitating a set of seed videos. In experiments, the auth... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.106038"
} | {
"id": "bFMBoHjeD0",
"metareview": "The paper demonstrates the zero-shot video imitation capabilities by training self-supervised autoregressive Transformers on video data. Such in-context capabilities enable solving unseen tasks by watching video demonstrations.\n\nThe rebuttal is well-accepted by most reviewers.... | {
"decision": "Accept (Poster)"
} |
wldwEhQ7cl | 2404.14280v1 | Robust Deep Equivariant Structure from Motion | {
"content": "## Abstract\n\nAbstract Multiview Structure from Motion is a fundamental and challenging computer vision problem. A recent deep-based approach was proposed utilizing matrix equivariant architectures for the simultaneous recovery of camera pose and 3D scene structure from large image collections. This wo... | [
{
"id": "tpWvfmAK8v",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents an architecture for Multiview Structure from Motion (SfM) focusing on the robust recovery of camera pose and 3D scene structure from large, unc... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "1;2;3",
"contribution_avg": 2,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.106862"
} | {
"id": "Ex5FrcjC79",
"metareview": "## Summary\nThe paper presents a new architecture for Multiview Structure from Motion (SfM) that improves the recovery of camera pose and 3D scene structure from large, uncontrolled image collections with outliers. The architecture incorporates an outlier classification module a... | {
"decision": "Accept (Poster)"
} |
wmFp2aMhi0 | 2410.21072v1 | Federated Time Series Generation on Feature and Temporally Misaligned Data | {
"content": "## Abstract\n\nAbstract Distributed time series data presents a challenge for federated learning, as clients often possess different feature sets and have misaligned time steps. Existing federated time series models are limited by the assumption of perfect temporal or feature alignment across clients. I... | [
{
"id": "hfs1FqpWX0",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper addresses the challenge of synthesizing time series data in a federated context, where the time series data at the clients may be misaligned either in t... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;5;3;4",
"confidence_avg": 4.25,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.107490"
} | {
"id": "EY5iOr7ct0",
"metareview": "This paper presents a federated time series diffusion model that jointly learns a synthesizer across clients. Although the reviewers liked the motivation, presentation, and good performance, they also raised many concerns on the experiment setting and results. The reviewers we... | {
"decision": "Reject"
} |
wnT8bfJCDx | 2405.16504v2 | Explaining Modern Gated-Linear RNNs via a Unified Implicit Attention Formulation | {
"content": "## Abstract\n\nAbstract Recent advances in efficient sequence modeling have led to attention-free layers, such as Mamba, RWKV, and various gated RNNs, all featuring sub-quadratic complexity in sequence length and excellent scaling properties, enabling the construction of a new type of foundation models.... | [
{
"id": "FMgXdF5GXS",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper studies the problem of sequence modelling. The authors aim to provide a unified framework of the recent attention-free methods such as Mamba and RWKV. T... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;2;2;3",
"confidence_avg": 2.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.108413"
} | {
"id": "mUMc6RgZkL",
"metareview": "This paper provides a unified view of attention-free architectures as implicit causal self-attention layers, which they apply to explainability. They also show good results on semantic segmentation and in-context learning.\n\nOverall, the reviewer's side on the acceptance. The m... | {
"decision": "Accept (Poster)"
} |
wozhdnRCtw | 2410.12877v1 | Improving Instruction-Following in Language Models through Activation Steering | {
"content": "## Abstract\n\nAbstract The ability to follow instructions is crucial for numerous real-world applications of language models. In pursuit of deeper insights and more powerful capabilities, we derive instruction-specific vector representations from language models and use them to steer models accordingly... | [
{
"id": "hWYg6ZYXIj",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a novel method to improve the instruction-following capabilities of language models using activation steering. Experiments on four different l... | {
"rating": "6;6;6;6",
"rating_avg": 6,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;2;3;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": "2026-01-14T22:16:04.109164"
} | {
"id": "xaMaAfz7ob",
"metareview": "## Summary\nThis paper proposes a method to enhance instruction-following in language models using activation steering. The proposed method computes steering vectors by contrasting model activations from inputs with and without instructions. These vectors are used at inference t... | {
"decision": "Accept (Poster)"
} |
wqA7QmpUwa | 2409.02889v2 | LongLLaVA: Scaling Multi-modal LLMs to 1000 Images Efficiently via a Hybrid Architecture | {
"content": "## Abstract\n\nAbstract Expanding the long-context capabilities of Multi-modal Large Language Models (MLLMs) is crucial for video understanding, high-resolution image understanding, and multi-modal agents. This involves a series of systematic optimizations, including model architecture, data constructio... | [
{
"id": "r57AyHYnbT",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents LongLLaVA, a novel solution to enhance the long-context capabilities. Architecture-wise, it combines Mamba (pseudo attention) and token compres... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "3;5;5;2",
"confidence_avg": 3.75,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "3;2;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": "2026-01-14T22:16:04.110244"
} | {
"id": "N7An1sTHdN",
"metareview": "Summary: The paper introduces LongLLaVA, a novel multimodal large language model designed to enhance long-context capabilities. It employs a hybrid architecture combining Mamba and Transformer layers and uses token compression to efficiently process a large number of images. Lon... | {
"decision": "Reject"
} |
wrXCIsysqB | 2410.01535v2 | GaussianBlock: Building Part-Aware Compositional and Editable 3D Scene by Primitives and Gaussians | {
"content": "## Abstract\n\nAbstract Recently, with the development of Neural Radiance Fields and Gaussian Splatting, 3D reconstruction techniques have achieved remarkably high fidelity. However, the latent representations learnt by these methods are highly entangled and lack interpretability. In this paper, we prop... | [
{
"id": "oIvlTLZOkA",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper is on 3D part aware semantic editing of scenes using Gaussian Splatting . Similar to\nprevious work like GaussianAvatar the paper uses a prior for initi... | {
"rating": "5;5;5;6;6",
"rating_avg": 5.4,
"confidence": "4;4;4;4;2",
"confidence_avg": 3.6,
"soundness": "2;2;3;3;3",
"soundness_avg": 2.6,
"contribution": "3;3;3;3;3",
"contribution_avg": 3,
"presentation": "3;3;2;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.110981"
} | {
"id": "mUstEAyikm",
"metareview": "This paper introduced GaussianBlock, a hybrid 3D representation combining primitives and 3D Gaussians for unsupervised, part-aware scene decomposition guided by 2D semantic priors. It claims to achieve disentangled, editable, high-fidelity reconstructions without requiring groun... | {
"decision": "Accept (Poster)"
} |
wryFCrWB0A | 2410.01912v1 | A Spark of Vision-Language Intelligence: 2-Dimensional Autoregressive Transformer for Efficient Finegrained Image Generation | {
"content": "## Abstract\n\nAbstract This work tackles the information loss bottleneck of vector-quantization (VQ) autoregressive image generation by introducing a novel model architecture called the 2-Dimensional Autoregression (DnD) Transformer. The DnD-Transformer predicts more codes for an image by introducing a... | [
{
"id": "6CquuaOS99",
"initial_rating": 6,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This paper addresses the problem of image modeling and generation using discrete latent representations and autoregressive (AR) sampling. The authors note that la... | {
"rating": "3;5;5;6;6",
"rating_avg": 5,
"confidence": "4;4;4;4;3",
"confidence_avg": 3.8,
"soundness": "1;3;2;2;4",
"soundness_avg": 2.4,
"contribution": "1;2;3;3;3",
"contribution_avg": 2.4,
"presentation": "1;3;3;3;4",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.111680"
} | {
"id": "yUqcAKdV5L",
"metareview": "The paper introduces a new formulation for vector-quantized autoregressive image generation that overcomes complications of existing 1D approaches (namely information loss and computational complexity) by predicting tokens that are expanded depth-wise. This gives rise to a 2D se... | {
"decision": "Accept (Poster)"
} |
ws5phQki00 | 2406.12480v1 | The Power of LLM-Generated Synthetic Data for Stance Detection in Online Political Discussions | {
"content": "## Abstract\n\nAbstract Stance detection holds great potential for enhancing the quality of online political discussions, as it\nhas shown to be useful for summarizing discussions, detecting misinformation, and evaluating opinion distributions.\nUsually, transformer-based models are used directly for st... | [
{
"id": "AbnrOen91J",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper tries to improve transformer-based stance detection models by fine-tuning on LLM generated data. They compare the real-world data with the synthetic dat... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "2;3;3",
"confidence_avg": 2.6666666666666665,
"soundness": "3;2;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;3;2",
"presentation_avg": 2.33333333333333... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.112372"
} | {
"id": "GwxN917QM2",
"metareview": "**Summary:**\n\nThe authors explore the efficacy of using LLM-generated synthetic training data for automated stance detection. They primarily test against a Bert baseline where individual models are trained for different stance detection question types. Varying levels of synthe... | {
"decision": "Accept (Spotlight)"
} |
wsWCVrH9dv | 2410.22598v1 | Feature Responsiveness Scores: Model-Agnostic Explanations for Recourse | {
"content": "## Abstract\n\nAbstract Machine learning models are often used to automate or support decisions in applications such as lending and hiring.\nIn such settings, consumer protection rules mandate that we provide a list of “principal reasons” to consumers who receive adverse decisions.\nIn practice, lenders... | [
{
"id": "fiJ24ksXj3",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper addresses the limitations of feature attribution methods in decision-making scenarios where regulations mandate that individuals are informed of the pr... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;4;3;5",
"confidence_avg": 4,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;2;4",
"contribution_avg": 2.5,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.113122"
} | {
"id": "KQzMfb5Zaj",
"metareview": "This paper sits broadly in the XAI space, more specifically the recourse portion of XAI - recourse roughly meaning a set of steps an input might take to adjust its features such that it crosses a decision boundary from a negative outcome to a positive one. Reviewers appreciated... | {
"decision": "Accept (Poster)"
} |
wsb9GNh1Oi | 2411.02158v1 | Learning Multiple Initial Solutions to Optimization Problems | {
"content": "## Abstract\n\nAbstract Sequentially solving similar optimization problems under strict runtime constraints is essential for many applications, such as robot control, autonomous driving, and portfolio management. The performance of local optimization methods in these settings is sensitive to the initial... | [
{
"id": "012pCihSpu",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces a method for learning a set of candidate initial solutions to warm-start optimal control problems. It proposes a series of objectives that en... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "5;5;3;3",
"confidence_avg": 4,
"soundness": "4;2;3;2",
"soundness_avg": 2.75,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"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": "2026-01-14T22:16:04.113987"
} | {
"id": "rsDtfQ1zhd",
"metareview": "This paper presented a novel framework for predicting multiple initial solutions for sequential black-box optimization problems.\nThe authors introduced two strategies to predict initial solutions called single-optimizer and multiple-optimizers (not necessarily running in parall... | {
"decision": "Reject"
} |
wwO8qS9tQl | 2312.12747v1 | ALMANACS: A Simulatability Benchmark for Language Model Explainability | {
"content": "## Abstract\n\nAbstract How do we measure the efficacy of language model explainability methods?\nWhile many explainability methods have been developed, they are typically evaluated on bespoke tasks, preventing an apples-to-apples comparison. To help fill this gap, we present ALMANACS, a language model ... | [
{
"id": "Zol4UaY678",
"initial_rating": 3,
"confidence": 4,
"soundness": 1,
"contribution": 2,
"presentation": 3,
"summary": "The authors propose ALMANACS: a benchmark that can be used to evaluate the extent to which different language model explanation methods can aid simulatability (th... | {
"rating": "3;3;3;3",
"rating_avg": 3,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;2;1",
"soundness_avg": 2.25,
"contribution": "2;2;1;2",
"contribution_avg": 1.75,
"presentation": "3;2;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.115416"
} | {
"id": "vH7bYYBVdU",
"metareview": "PC is entering meta-review on behalf of SAC and AC:\n\n The reviewers felt that the submission was not a strong contribution, and the authors did not respond during the review proicess.",
"additional_comments": "TBD"
} | {
"decision": "Reject"
} |
wwVGZRnAYG | 2410.20971v1 | BlueSuffix: Reinforced Blue Teaming for Vision-Language Models Against Jailbreak Attacks | {
"content": "## Abstract\n\nAbstract Despite their superb multimodal capabilities, Vision-Language Models (VLMs) have been shown to be vulnerable to jailbreak attacks, which are inference-time attacks that induce the model to output harmful responses with tricky prompts. It is thus essential to defend VLMs against p... | [
{
"id": "eIWGLA1zUA",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes a defense method for jailbreak attacks on VLMs to protect VLMs from black-box jailbreak attacks. Specifically, the method utilizes a multimoda... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "3;3;5;5",
"confidence_avg": 4,
"soundness": "3;3;3;2",
"soundness_avg": 2.75,
"contribution": "3;2;2;2",
"contribution_avg": 2.25,
"presentation": "3;3;3;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.116422"
} | {
"id": "NDAO7f8jLH",
"metareview": "The recommendation is based on the reviewers' comments, the area chair's evaluation, and the author-reviewer discussion. \n\nThis paper proposes a blue-team method (BlueSuffix) for vision language models, through purification techniques for different modalities. All reviewers fi... | {
"decision": "Accept (Poster)"
} |
wy9FRV8O5s | 2406.02929v1 | ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning | {
"content": "## Abstract\n\nAbstract Zero-Shot Learning (ZSL) aims to enable classifiers to identify unseen classes by enhancing data efficiency at the class level. This is achieved by generating image features from pre-defined semantics of unseen classes. However, most current approaches heavily depend on the numbe... | [
{
"id": "TZ57ToivvT",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes to exploit diffusion mechanism to enhance the generative models in zero-shot learning. Specifically, existing generative zero-shot learning me... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "5;5;4",
"confidence_avg": 4.666666666666667,
"soundness": "3;3;4",
"soundness_avg": 3.3333333333333335,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.117043"
} | {
"id": "vBstLEgMoH",
"metareview": "# Summary and Recommendation for Acceptance\n\n---\n## **Strengths**\n1. **Novel Contributions**:\n - Proposes **ZeroDiff**, a generative framework addressing spurious visual-semantic correlations in zero-shot learning (ZSL). Key innovations include:\n - **Diffusion augmen... | {
"decision": "Accept (Poster)"
} |
x1An5a3U9I | 2406.09357v2 | Advancing Graph Generation through Beta Diffusion | {
"content": "## Abstract\n\nAbstract Diffusion models have excelled in generating natural images and are now being adapted to a variety of data types, including graphs. However, conventional models often rely on Gaussian or categorical diffusion processes, which can struggle to accommodate the mixed discrete and con... | [
{
"id": "RCBJW1SeOA",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper proposes graph beta diffusion for the task of generating small graphs such as molecules. The core concept is approaching this task by adapting beta diff... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;3;2;5",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "4;2;2;3",
"contribution_avg": 2.75,
"presentation": "3;3;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.117744"
} | {
"id": "TNNx9v0oQu",
"metareview": "This paper introduces Graph Beta Diffusion (GBD), a generative model designed to handle the unique mix of discrete and continuous components in graph data. By using a beta diffusion process and a modulation technique, GBD generates realistic graphs and outperforms existing model... | {
"decision": "Accept (Poster)"
} |
x1Okv4kbVR | 2410.07672v1 | MACPO: Weak-to-Strong Alignment via Multi-Agent Contrastive Preference Optimization | {
"content": "## Abstract\n\nAbstract As large language models (LLMs) are rapidly advancing and achieving near-human capabilities, aligning them with human values is becoming more urgent.\nIn scenarios where LLMs outperform humans, we face a weak-to-strong alignment problem where we need to effectively align strong s... | [
{
"id": "gYAPFYhQ3x",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 2,
"summary": "The paper explores the alignment problem when the LLM outperforms humans and human supervision is therefore weak. The authors propose a multiagent contrastive pre... | {
"rating": "3;6;6;6",
"rating_avg": 5.25,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;1;3;2",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.118517"
} | {
"id": "Tdu65pJvr2",
"metareview": "The paper proposes Multi-Agent Contrastive Preference Optimization (MACPO) to encourage strong students and weak teachers to learn from each other through contrastive learning. All the reviewers found the paper easy to follow and are satisfied with the strong empirical results. ... | {
"decision": "Accept (Poster)"
} |
x1uv2gdjKV | 2405.18881v3 | Inference-Time Alignment of Diffusion Models with Direct Noise Optimization | {
"content": "## Abstract\n\nAbstract In this work, we focus on the alignment problem of diffusion models with a continuous reward function, which represents specific objectives for downstream tasks, such as increasing darkness or improving the aesthetics of images. The central goal of the alignment problem is to adj... | [
{
"id": "falzE357lA",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a method called direct noise optimization (DNO) for maximizing some reward function for the generated samples. Different from finetuning and... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "3;3;2;4",
"confidence_avg": 3,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.119511"
} | {
"id": "DatLle1mG3",
"metareview": "This paper presents a method called *Direct Noise Optimization* (DNO) for maximizing a reward function associated with generated samples. Unlike finetuning or reinforcement learning approaches, DNO functions as a test-time optimization technique. The method is further extended t... | {
"decision": "Reject"
} |
x1yOHtFfDh | 2410.08474v2 | SportU: A Comprehensive Sports Understanding Benchmark for Multimodal Large Language Models | {
"content": "## Abstract\n\nAbstract Multimodal Large Language Models (MLLMs) are advancing the ability to reason about complex sports scenarios by integrating textual and visual information. To comprehensively evaluate their capabilities, we introduce SPORTU, a benchmark designed to assess MLLMs across multi-level ... | [
{
"id": "oIUOcvFoLE",
"initial_rating": 5,
"confidence": 4,
"soundness": 1,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces SPORTU, a new benchmark designed to evaluate the capabilities of Multimodal Large Language Models (MLLMs) in sports understanding and reason... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;1;4;2",
"soundness_avg": 2.25,
"contribution": "2;2;4;2",
"contribution_avg": 2.5,
"presentation": "2;2;4;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.120455"
} | {
"id": "ojM0U2gWIB",
"metareview": "Summary: This paper introduced a benchmark to comprehensively evaluate the capabilities of MLLMs, which include multi-level sports reasoning tasks. Multiple baselines are considered for comprehensive evaluations.\n\nStrengths: (1) The proposed benchmark could be helpful for furt... | {
"decision": "Accept (Poster)"
} |
x4ZmQaumRg | 2408.01536v1 | Active Learning for Neural PDE Solvers | {
"content": "## Abstract\n\nAbstract Solving partial differential equations (PDEs) is a fundamental problem in engineering and science. While neural PDE solvers can be more efficient than established numerical solvers, they often require large amounts of training data that is costly to obtain. Active Learning (AL) c... | [
{
"id": "3ofFFxd1tn",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper provided a bechmark called AL4PDE, which unifies active learning (AL) with neural PDE solvers. Specifically, it studies how several state-of-the-art ne... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "3;3;4;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "1;2;3;3",
"contribution_avg": 2.25,
"presentation": "3;3;4;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.121237"
} | {
"id": "TcAGAD2qSL",
"metareview": "This paper proposes a benchmark study for neural PDE solvers with active learning approaches. It provides a modular benchmark with various parametric PDEs as well as active learning methods. During the rebuttal, the authors have addressed the reviewers' questions and concerns on... | {
"decision": "Accept (Poster)"
} |
x5YEibapUM | 2411.01663v1 | Unlocking the Theory Behind Scaling 1-Bit Neural Networks | {
"content": "## Abstract\n\nRecently, 1-bit Large Language Models (LLMs) have emerged, showcasing an impressive combination of efficiency and performance that rivals traditional LLMs. Research by WMD + [ 91 ], MWM + [ 70 ] indicates that the performance of these 1-bit LLMs progressively improves as the number of par... | [
{
"id": "3cLVOZQj6L",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper provides theoretical justification for scaling 1-bit neural networks, showing that their training dynamics converge to kernel-like behavior as model wi... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "3;4;3;4",
"confidence_avg": 3.5,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;1;3;2",
"contribution_avg": 2,
"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": "2026-01-14T22:16:04.122634"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
x5hXkSMOd1 | 2408.10202v1 | SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP | {
"content": "## Abstract\n\nAbstract Large-scale vision-language models, such as CLIP, are known to contain harmful societal bias regarding protected attributes ( e.g . , gender and age). In this paper, we aim to address the problems of societal bias in CLIP. Although previous studies have proposed to debias societa... | [
{
"id": "XAAoqOzOKI",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper aims to overcome societal bias present in datasets used to train large scale multi-modal models like CLIP. The authors provide a study of debiasing met... | {
"rating": "5;5;8;8",
"rating_avg": 6.5,
"confidence": "3;3;4;4",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;4;3",
"contribution_avg": 2.75,
"presentation": "3;2;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.124049"
} | {
"id": "YVfhyrMUj9",
"metareview": "This paper focuses on the societal bias issue in large-scale vision-language models and conducts a study of debiasing methods to present the limitations of existing methods. To overcome the societal biases, the authors propose a simple yet effective debiasing method SANER to inc... | {
"decision": "Accept (Poster)"
} |
x6YSsKYJuH | 2404.19597v2 | TuBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning | {
"content": "## Abstract\n\nAbstract The implications of backdoor attacks on English-centric large language models (LLMs) have been widely examined — such attacks can be achieved by embedding malicious behaviors during training and activated under specific conditions that trigger malicious outputs.\nDespite the incr... | [
{
"id": "AomFJgGjWx",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper leverages instruction tuning and explores the backdoor transfer abilities of large language models across multiple languages. The paper empirically ana... | {
"rating": "5;5;5",
"rating_avg": 5,
"confidence": "5;3;4",
"confidence_avg": 4,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "3;3;2",
"contribution_avg": 2.6666666666666665,
"presentation": "3;3;2",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.124904"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
x83w6yGIWb | 2410.17711v1 | Beware of Calibration Data for Pruning Large Language Models | {
"content": "## Abstract\n\nAbstract As large language models (LLMs) are widely applied across various fields, model compression has become increasingly crucial for reducing costs and improving inference efficiency.\nPost-training pruning is a promising method that does not require resource-intensive iterative train... | [
{
"id": "Zhv6PjSvmC",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper investigates the role of calibration data in post-training pruning for large language models (LLMs). The authors find that calibration data similar to ... | {
"rating": "1;5;6;8",
"rating_avg": 5,
"confidence": "5;3;3;5",
"confidence_avg": 4,
"soundness": "1;3;3;4",
"soundness_avg": 2.75,
"contribution": "1;3;2;3",
"contribution_avg": 2.25,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.126066"
} | {
"id": "579ZmYMHnU",
"metareview": "Post-hoc pruning methods are a common and efficient approach for compressing pretraining language models; however, these approaches rely on some auxiliary data, or \"calibration data,\" to estimate parameter importance. This work tests how the choice of calibration data affects ... | {
"decision": "Accept (Poster)"
} |
xAM9VaXZnY | 2406.05815v2 | What Can We Learn from State Space Models for Machine Learning on Graphs? | {
"content": "## Abstract\n\nAbstract Machine learning on graphs has recently found extensive applications across domains. However, the commonly used Message Passing Neural Networks (MPNNs) suffer from limited expressive power and struggle to capture long-range dependencies. Graph transformers offer a strong alternat... | [
{
"id": "JjP7QkqWRV",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This work extends the state space models (SSMs) from sequence modeling to the domain of graph-structured data. By tailoring SSMs for graphs, the proposed model (... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;2;2;2",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.127107"
} | {
"id": "QmwZmGrvoT",
"metareview": "The paper does not receive consistently positive support from the reviewers even though after extensive discussions.",
"additional_comments": "Most of the major issues remain unsolved. One of the reviewers was willing to raise the score slightly after discussions while other r... | {
"decision": "Reject"
} |
xCFdAN5DY3 | 2301.10343v5 | A Foundation Model for Weather and Climate | {
"content": "## Abstract\n\nAbstract Most state-of-the-art approaches for weather and climate modeling are based on physics-informed numerical models of the atmosphere. These approaches aim to model the non-linear dynamics and complex interactions between multiple variables, which are challenging to approximate. Add... | [
{
"id": "jNIdr7btzq",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a new foundation model, Prithvi WxC, for atmospheric modeling applications in weather and climate. Prithvi WxC was trained on 3-hourly data... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"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": "2026-01-14T22:16:04.128170"
} | {
"id": "u3W7l5F42a",
"metareview": "The paper proposes a foundation model for weather and climate. While the foundation model backbone is similar to many other works in the realm of foundation models, the paper proposes interesting objectives based on the weather domain. The biggest weakness of the paper is in exp... | {
"decision": "Reject"
} |
xCMmtYOsiL | 2411.04491v1 | Series-to-Series Diffusion Bridge Model | {
"content": "## Abstract\n\nAbstract Diffusion models have risen to prominence in time series forecasting, showcasing their robust capability to model complex data distributions. However, their effectiveness in deterministic predictions is often constrained by instability arising from their inherent stochasticity. I... | [
{
"id": "AItS93OZnf",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents a comprehensive framework that encompasses most existing diffusion-based methods. Building on this foundation, the authors introduce the Serie... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;2;3",
"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": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.129098"
} | {
"id": "uJ1gZPcaAa",
"metareview": "The paper introduces a unified framework for diffusion models for time series forecasting and, building on this framework, it introduces the Brownian bridge process to enhance prediction accuracy. In terms of overall ratings, one reviewer is positive, two reviewers are mildly po... | {
"decision": "Reject"
} |
xCkgX4Xfu0 | 2408.05804v1 | A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals | {
"content": "## Abstract\n\nAbstract In this paper, we present empirical evidence of skills and directed exploration emerging from a simple RL algorithm long before any successful trials are observed. For example, in a manipulation task, the agent is given a single observation of the goal state (see Fig. 1 ) and lea... | [
{
"id": "QXeLxc2UHK",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This work aims to solve goal-conditioned RL (GCRL) tasks where only one goal is desired without access to rewards (dense or sparse), curriculums, or demonstration... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;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": "2026-01-14T22:16:04.130065"
} | {
"id": "o60h8uV9VT",
"metareview": "This paper presents an interesting phenomena in contrastive RL where a single hard goal lead to better performing policies than learning with curricula. Empirical results show that diverse skills emerge during learning long before the agent reaches the goal state and receives a ... | {
"decision": "Accept (Poster)"
} |
xFvHcgj1fO | 2409.09742v1 | OML-AD: Online Machine Learning for Anomaly Detection in Time Series Data | {
"content": "## Abstract\n\nAbstract Time series are ubiquitous and occur naturally in a variety of applications – from data recorded by sensors in manufacturing processes, over financial data streams to climate data. Different tasks arise, such as regression, classification or segmentation of the time series. Howev... | [
{
"id": "4Syfcyzz1Y",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 2,
"summary": "The paper presents OML-AD, an online anomaly detection method for non-stationary time-series data. The method combines existing ideas scatter in literature for di... | {
"rating": "3;3;3;3",
"rating_avg": 3,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "1;1;1;2",
"soundness_avg": 1.25,
"contribution": "2;1;1;1",
"contribution_avg": 1.25,
"presentation": "1;2;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.131317"
} | {
"id": "U4VSLzXFYX",
"metareview": "The paper presents OML-AD, an online method for anomaly detection that aims to handle non-stationary time series by adapting a forecasting model on the fly. While it tries to address the concept drift problem, the actual improvements shown over existing baselines seem pretty mod... | {
"decision": "Reject"
} |
xH53mFbwK8 | 2407.04108v2 | Future Events as Backdoor Triggers: Investigating Temporal Vulnerability in LLMs | {
"content": "## Abstract\n\nAbstract Backdoors are hidden behaviors that are only triggered once an AI system has been deployed. Bad actors looking to create successful backdoors must design them to avoid activation during training and evaluation. Since data used in these stages often only contains information about... | [
{
"id": "mMOfaltRb4",
"initial_rating": 3,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper studies the problem of deceptive alignment, where an AI system behaves differently in evaluation and deployment. In particular, building on Hubinger et... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "3;3;3;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.131972"
} | {
"id": "z9yTAXDtr0",
"metareview": "The paper investigates strategic deception in LLMs via temporal backdoors that activate on future events. The strengths include: demonstrating LLMs can distinguish past/future events with 90% accuracy, successfully implementing temporal backdoors, and showing HHH fine-tuning hel... | {
"decision": "Reject"
} |
xHGL9XqR8Y | 2406.12179v1 | The Wisdom of a Crowd of Brains: A Universal Brain Encoder | {
"content": "## Abstract\n\nAbstract Image-to-fMRI encoding is important for both neuroscience research and practical applications.\nHowever, such “Brain-Encoders” have been typically trained per-subject and per fMRI-dataset,\nthus restricted to very limited training data.\nIn this paper we propose a Universal Brain... | [
{
"id": "CBUP6mhKxm",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a Universal fMRI Encoder for the prediction of brain responses to image stimuli. Unlike traditional subject-specific brain encoding models, ... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "4;5;4;3",
"confidence_avg": 4,
"soundness": "1;2;3;4",
"soundness_avg": 2.5,
"contribution": "1;2;3;4",
"contribution_avg": 2.5,
"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": "2026-01-14T22:16:04.132674"
} | {
"id": "Yc9q0i4q0h",
"metareview": "This manuscript proposes a model for the prediction of volumes of fMRI (BOLD) activations given images, presumably image stimuli. It claims generalization (\"universal\" encoder) allowing for predictions to be made in a variety of settings in a variety of subjects. The authors d... | {
"decision": "Reject"
} |
xIUUnzrUtD | 2410.21332v1 | Building, Reusing, and Generalizing Abstract Representations from Concrete Sequences | {
"content": "## Abstract\n\nAbstract Humans excel at learning abstract patterns across different sequences, filtering out irrelevant details, and transferring these generalized concepts to new sequences.\nIn contrast, many sequence learning\nmodels lack the ability to abstract, which leads to memory\ninefficiency an... | [
{
"id": "qCMUdPUNfd",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "**Update after rebuttal:** Thanks to the very detailed response by the authors, my questions and potential misunderstandings have been clarified and I have raised... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "3;4;4;2",
"confidence_avg": 3.25,
"soundness": "2;4;4;3",
"soundness_avg": 3.25,
"contribution": "2;4;4;3",
"contribution_avg": 3.25,
"presentation": "3;2;4;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.133743"
} | {
"id": "zQVTf5Ch9U",
"metareview": "This paper introduces the Hierarchical Variable Learning Model (HVM), a novel extension of the Hierarchical Chunking Model (HCM) that learns abstract representations by grouping sequence chunks into higher-level categories based on contextual similarity. This enables HVM to achi... | {
"decision": "Accept (Poster)"
} |
xImTb8mNOr | 2406.11463v1 | Just How Flexible are Neural Networks in Practice? | {
"content": "## Abstract\n\nAbstract It is widely believed that a neural network can fit a training set containing at least as many samples as it has parameters, underpinning notions of overparameterized and underparameterized models. In practice, however, we only find solutions accessible via our training procedure... | [
{
"id": "zfJrYqvmlO",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper empirically investigates the practical flexibility and capacity of neural networks to fit data, introducing several key findings:\n\n1. Practical Capac... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "3;4;3;3;4",
"confidence_avg": 3.4,
"soundness": "2;3;3;3;3",
"soundness_avg": 2.8,
"contribution": "2;1;2;2;2",
"contribution_avg": 1.8,
"presentation": "2;3;3;2;4",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.134528"
} | {
"id": "DHLNMxzX7H",
"metareview": "The Effective Model Complexity (EMC) metric is used as a predictor of generalization performance. The paper examines the practical limitations of neural networks, showing that standard optimizers often do not achieve theoretical capacity. It finds that CNNs are more parameter-ef... | {
"decision": "Reject"
} |
xJc3PazBwS | 2410.03037v1 | Disentangling Textual and Acoustic Features of Neural Speech Representations | {
"content": "## Abstract\n\nAbstract Neural speech models build deeply entangled internal representations, which capture a variety of features (e.g., fundamental frequency, loudness, syntactic category, or semantic content of a word) in a distributed encoding.\nThis complexity makes it difficult to track the extent ... | [
{
"id": "dUsbNyZc51",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "Many standard speech representations are learned in a self-supervised way (HuBERT, w2v2, etc) and hence are, essentially, entangled blackboxes that have acoustic ... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;3;2;2",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.135223"
} | {
"id": "SHN32KJ9uw",
"metareview": "This paper proposes to use VIB (Variational Information Bottleneck) as a framework for disentangling speech representations from neural speech models (like Wav2Vec2 and HuBERT) into two distinct components: textual content (what can be transcribed as text) and acoustic features ... | {
"decision": "Reject"
} |
xJljiPE6dg | 2409.12822v2 | Language Models Learn to Mislead Humans via RLHF | {
"content": "## Abstract\n\nAbstract Language models (LMs) can produce errors that are hard to detect for humans, especially when the task is complex.\nRLHF, the most popular post-training method, may exacerbate this problem: to achieve higher rewards, LMs might get better at convincing humans that they are right ev... | [
{
"id": "Q9cO6xgRg2",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper \"Language Models Learn to Mislead Humans via RLHF\" examines how language models fine-tuned with Reinforcement Learning from Human Feedback (RLHF) can... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;4;3;4",
"contribution_avg": 3.25,
"presentation": "3;4;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.135922"
} | {
"id": "MHnfuaGrmk",
"metareview": "The paper studies the phenomenon of “U-SOPHISTRY” where LLMs trained with RLHF unintentionally become more convincing to humans about incorrect outputs without improving actual performance. The work performs extensive experiments on QA and programming tasks, and shows that RLHF-... | {
"decision": "Accept (Poster)"
} |
xNgmEWmd9T | 2409.17092v1 | Accumulator-Aware Post-Training Quantization for Large Language Models | {
"content": "## Abstract\n\nAbstract Several recent studies have investigated low-precision accumulation, reporting improvements in throughput, power, and area across various platforms.\nHowever, the accompanying proposals have only considered the quantization-aware training (QAT) paradigm, in which models are fine-... | [
{
"id": "FGnCCkfuMj",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work investigates post training quantization from an accumulator-aware perspective. They aim at using low-precision at accumulation while avoiding the overfl... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "3;4;2;3",
"confidence_avg": 3,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"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": "2026-01-14T22:16:04.137019"
} | {
"id": "tVi9RJLj68",
"metareview": "While the problem addressed by the paper—extending post-training quantization (PTQ) to consider accumulator-aware design—is well-motivated, the work falls short in providing sufficient experimental evidence to validate its claims and situate its contributions within the current ... | {
"decision": "Reject"
} |
xNsIfzlefG | 2401.00036v2 | Discrete Distribution Networks | {
"content": "## Abstract\n\nAbstract We introduce a novel generative model, the Discrete Distribution Networks (DDN), that approximates data distribution using hierarchical discrete distributions. We posit that since the features within a network inherently capture distributional information, enabling the network to... | [
{
"id": "iX2q1x3Lbk",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes to generate samples in a sequential way: pass the output of the previous module into a function randomly selected from a set of K functions in ... | {
"rating": "3;6;8",
"rating_avg": 5.666666666666667,
"confidence": "4;4;4",
"confidence_avg": 4,
"soundness": "2;3;4",
"soundness_avg": 3,
"contribution": "2;3;4",
"contribution_avg": 3,
"presentation": "2;2;4",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.137716"
} | {
"id": "ZuK3Mlasfx",
"metareview": "This paper suggests the Discrete Distribution Network (DDN) as a new generative model paradigm. The DDN is composed of Discrete Distribution Layers (DDLs) that each generate several images out of its input feature. For training the closest generated image to the training sample ... | {
"decision": "Accept (Poster)"
} |
xOmC5LiVuN | 2411.02372v1 | Learning General-purpose Biomedical Volume Representations using Randomized Synthesis | {
"content": "## Abstract\n\nAbstract Current volumetric biomedical foundation models struggle to generalize as public 3D datasets are small and do not cover the broad diversity of medical procedures, conditions, anatomical regions, and imaging protocols. We address this by creating a representation learning method t... | [
{
"id": "hCejq4vkvv",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The work proposes a new method for general (few-shot) medical image segmentation and registration. The method is based on generating realistic and varying synthet... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.138478"
} | {
"id": "ipwKy3LDAF",
"metareview": "The submission presents a method for pre-training a generalist 3D backbone for medical image segmentation and registration using synthetic data, combining domain randomization and local contrastive learning. Reviewers praised the method's innovation, sustainability (fewer traina... | {
"decision": "Accept (Poster)"
} |
xP1radUi32 | 2410.01294v1 | Endless Jailbreaks with Bijection Learning | {
"content": "## Abstract\n\nAbstract Despite extensive safety training, LLMs are vulnerable to adversarial inputs. In this work, we introduce a simple but powerful attack paradigm, bijection learning , that yields a practically endless set of jailbreak prompts. We exploit language models’ advanced reasoning capabili... | [
{
"id": "Ay7nRu2PB1",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper discusses an approach to jailbreaking via bijection learning. Specifically, they generate a random transformation over characters that shifts the input ... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;2;2",
"contribution_avg": 2.25,
"presentation": "3;1;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.139214"
} | {
"id": "yL8sr1wjR7",
"metareview": "The paper proposes a novel bijection method leveraging side-channel techniques to jailbreak large language models (LLMs). The findings indicate that stronger LLMs are more vulnerable to jailbreak attacks and demonstrate state-of-the-art (SOTA) attack success rates. The proposed ... | {
"decision": "Accept (Poster)"
} |
xQBRrtQM8u | 2409.08861v4 | Adjoint Matching: Fine-tuning Flow and Diffusion Generative Models with Memoryless Stochastic Optimal Control | {
"content": "## Abstract\n\nAbstract Dynamical generative models that produce samples through an iterative process, such as Flow Matching and denoising diffusion models, have seen widespread use, but there has not been many theoretically-sound methods for improving these models with reward fine-tuning.\nIn this work... | [
{
"id": "2CC4rdfNQY",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 3,
"summary": "This paper provides theorectical insights on why optimizing a KL-regularized reward objective (which is popular and dominant in RLHF for LLM) could lead to a bias... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "4;4;3;4",
"soundness_avg": 3.75,
"contribution": "3;3;3;4",
"contribution_avg": 3.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.140407"
} | {
"id": "amAlozPLvi",
"metareview": "This paper addresses the challenge of reward fine-tuning in dynamical generative models, such as Flow Matching and denoising diffusion models, by framing it as a stochastic optimal control (SOC) problem. It introduces a theoretically grounded approach requiring a specific memory... | {
"decision": "Accept (Spotlight)"
} |
xQVxo9dSID | 2406.14548v2 | Consistency Models Made Easy | {
"content": "## Abstract\n\nAbstract Consistency models (CMs) offer faster sampling than traditional diffusion models, but their training is resource-intensive. For example, as of 2024, training a state-of-the-art CM on CIFAR-10 takes one week on 8 GPUs. In this work, we propose an effective scheme for training CMs ... | [
{
"id": "BRvRQiKQwZ",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 4,
"summary": "In this paper, the athors made the point that diffusion models can be viewed as a special case of CMs. Based on it, they fine-tuned a consistency model starting f... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "3;2;3;2",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;2;2;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.142749"
} | {
"id": "I34lq3sML2",
"metareview": "This paper explores a method to train consistency models by initializing them from pre-trained diffusion models and then fine-tuning. Various tricks are employed in the training to improve convergence and efficiency of the training, resulting in SOTA performance on small-scale i... | {
"decision": "Accept (Poster)"
} |
xSSo8kCA9G | 2411.07837v1 | FRUGAL: Memory-Efficient Optimization by Reducing State Overhead for Scalable Training | {
"content": "## Abstract\n\nAbstract With the increase in the number of parameters in large language models, the process of pre-training and fine-tuning increasingly demands larger volumes of GPU memory.\nA significant portion of this memory is typically consumed by the optimizer state.\nTo overcome this challenge, ... | [
{
"id": "tAned2tIdE",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The work proposes a memory efficient training method called FRUGAL which is essentially a combination of full-rank updates with gradient splitting. The authors pa... | {
"rating": "5;5;5;5;6",
"rating_avg": 5.2,
"confidence": "3;4;3;4;2",
"confidence_avg": 3.2,
"soundness": "3;3;3;2;3",
"soundness_avg": 2.8,
"contribution": "2;2;3;3;4",
"contribution_avg": 2.8,
"presentation": "2;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": "2026-01-14T22:16:04.147163"
} | {
"id": "deTxuN1K6F",
"metareview": "This paper proposes FRUGAL, a memory-efficient optimization framework that aims to provide full-rank updates for large language model (LLM) training by splitting parameters into those updated with stateful optimizers (e.g., Adam) and those updated with state-free methods (e.g., ... | {
"decision": "Reject"
} |
xUMI52rrW7 | 2410.02268v2 | Structural-Entropy-Based Sample Selection for Efficient and Effective Learning | {
"content": "## Abstract\n\nAbstract Sample selection improves the efficiency and effectiveness of machine learning models by providing informative and representative samples.\nTypically, samples can be modeled as a sample graph, where nodes are samples and edges represent their similarities.\nMost existing methods ... | [
{
"id": "S59e6prU5K",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper studies sample selection which aims to extract a small, representative subset from a larger dataset. The authors introduce a novel sample selection sche... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.148953"
} | {
"id": "7RAYlm0Pvw",
"metareview": "The reviewers raised several concerns during the review process regarding largely around lack of novelty and lack of theoretical justification. The authors have addressed most of these concerns. In my opinion the work is somewhat incremental, but the empirical results are intere... | {
"decision": "Accept (Poster)"
} |
xVefsBbG2O | 2410.02543v2 | Diffusion Models are Evolutionary Algorithms | {
"content": "## Abstract\n\nAbstract In a convergence of machine learning and biology, we reveal that diffusion models are evolutionary algorithms. By considering evolution as a denoising process and reversed evolution as diffusion, we mathematically demonstrate that diffusion models inherently perform evolutionary ... | [
{
"id": "7xqaltkJqF",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The authors try to connect Diffusion Models and Evolutionary computation by arguing that both processes do iterative refinements through an update rule plus some ... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"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": "2026-01-14T22:16:04.149945"
} | {
"id": "PVHZQLLhzi",
"metareview": "This paper bridges the fields of machine learning and evolutionary computation by revealing a mathematical equivalence between diffusion models and evolutionary algorithms. Diffusion models, initially designed for generative tasks, are reinterpreted as performing evolutionary al... | {
"decision": "Accept (Poster)"
} |
xW4J2QlqRx | 2410.12672v2 | Context Matters: Leveraging Contextual Features for Time Series Forecasting | {
"content": "## Abstract\n\nAbstract Time series forecasts are often influenced by exogenous contextual features in addition to their corresponding history. For example, in financial settings, it is hard to accurately predict a stock price without considering public sentiments and policy decisions in the form of new... | [
{
"id": "0OmZmszh0u",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 1,
"presentation": 2,
"summary": "This paper presents a work on aggregating contextual information into time series forecasting. Specifically, the authors propose to use a universal context encode... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "4;3;5;4",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;1;1",
"contribution_avg": 1.5,
"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": "2026-01-14T22:16:04.150664"
} | {
"id": "0pYHY9htpa",
"metareview": "In this paper, the authors propose an approach to leveraging contextual features in time series prediction. Reviewers found the paper well-written but noted that its contribution is limited, the theoretical analysis lacks depth, and the experimental results are insufficient. The... | {
"decision": "Reject"
} |
xXTkbTBmqq | 2409.02060v1 | OLMoE: Open Mixture-of-Experts Language Models | {
"content": "## Abstract\n\nAbstract We introduce OLMoE , a fully open, state-of-the-art language model leveraging sparse Mixture-of-Experts (MoE). OLMoE-1B-7B has 7 billion (B) parameters but uses only 1B per input token. We pretrain it on 5 trillion tokens and further adapt it to create OLMoE-1B-7B-Instruct . Our ... | [
{
"id": "wsaqtgZmFV",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "This paper introduces OLMoE, a fully open, state-of-the-art language model built on a sparse Mixture-of-Experts (MoE) architecture. The authors conducted extensiv... | {
"rating": "8;8;10",
"rating_avg": 8.666666666666666,
"confidence": "2;3;5",
"confidence_avg": 3.3333333333333335,
"soundness": "4;4;4",
"soundness_avg": 4,
"contribution": "3;4;4",
"contribution_avg": 3.6666666666666665,
"presentation": "4;4;4",
"presentation_avg": 4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.151798"
} | {
"id": "QlsXdHXmHv",
"metareview": "The paper introduces a state-of-the-art language model leveraging a sparse Mixture-of-Experts (MoE) architecture. It presents novel findings on MoE training, defines and analyzes new routing properties showing high specialization in their model, and open-sources all aspects of t... | {
"decision": "Accept (Oral)"
} |
xYzOkOGD96 | 2411.07584v1 | Grounded Video Caption Generation | {
"content": "## Abstract\n\nAbstract We propose a new task, dataset and model for grounded video caption generation.\nThis task unifies captioning and object grounding in video, where the objects in the caption are grounded in the video via temporally consistent bounding boxes. We introduce the following contributio... | [
{
"id": "wIum2rKzDC",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposed a new task (GROC), a manually annotated test dataset, and an automatically generated dataset. It introduces VideoGLaMM, a model designed to gen... | {
"rating": "3;3;3;3;5;6",
"rating_avg": 3.8333333333333335,
"confidence": "3;3;4;5;4;4",
"confidence_avg": 3.8333333333333335,
"soundness": "2;2;2;2;2;2",
"soundness_avg": 2,
"contribution": "3;2;2;2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "2;2;1;2;4;3",
"presentation_avg": 2.... | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.152878"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xZ2lTzfyFv | 2410.04196v1 | Improving Generalization with Flat Hilbert Bayesian Inference | {
"content": "## Abstract\n\nAbstract We introduce Flat Hilbert Bayesian Inference (FHBI), an algorithm designed to enhance generalization in Bayesian inference. Our approach involves an iterative two-step procedure with an adversarial functional perturbation step and a functional descent step within the reproducing ... | [
{
"id": "33uVeQ1nP0",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work presents an algorithm called Flat Hilbert Bayesian Inference (FHBI), which incorporates Sharpness-aware minimization (SAM) technique in Bayesian inferen... | {
"rating": "3;6;8;8",
"rating_avg": 6.25,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.153655"
} | {
"id": "ck8HkFTVZB",
"metareview": "This paper proposes Flat Hilbert Bayesian Inference (FHBI), an algorithm to enhance generalization in Bayesian inference. Similar to Sharpness-Aware Minimization (SAM), each iteration of FHBI involves an adversarial perturbation step followed by a descent step, with the key diff... | {
"decision": "Reject"
} |
xaYlO03tIk | 2411.04919v1 | Stem-OB: Generalizable Visual Imitation Learning with Stem-Like Convergent Observation through Diffusion Inversion | {
"content": "## Abstract\n\nAbstract Visual imitation learning methods demonstrate strong performance, yet they lack generalization when faced with visual input perturbations like variations in lighting and textures. This limitation hampers their practical application in real-world settings. To address this, we prop... | [
{
"id": "Vzchmw7qlc",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper introduces Stem-OB, a method that enhances visual imitation learning by using image inversion from pretrained diffusion models to reduce low-level visu... | {
"rating": "3;6;6;6",
"rating_avg": 5.25,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "1;3;3;3",
"soundness_avg": 2.5,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "2;3;4;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.154524"
} | {
"id": "qKwJhWV1yK",
"metareview": "This paper proposes Stem-OB, which suppresses low-level visual differences while maintaining high-level scene structures. The idea is to apply a diffusion inversion process to transform images into a shared image space without additional training. The approach is demonstrated wi... | {
"decision": "Accept (Spotlight)"
} |
xam3sR3ffY | 2406.12624v4 | Judging the Judges: Evaluating Alignment and Vulnerabilities in LLMs-as-Judges | {
"content": "## Abstract\n\nAbstract Offering a promising solution to the scalability challenges associated with human evaluation, the LLM-as-a-judge paradigm is rapidly gaining traction as an approach to evaluating large language models (LLMs).\nHowever, there are still many open questions about the strengths and w... | [
{
"id": "5PpgAujGJv",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This work provides an examination of LLM judges regarding their performance and vulnerabilities in a reference-based evaluation setting for QA tasks. Using human ... | {
"rating": "3;3;3;5;8",
"rating_avg": 4.4,
"confidence": "4;3;4;4;4",
"confidence_avg": 3.8,
"soundness": "1;2;2;3;4",
"soundness_avg": 2.4,
"contribution": "1;2;2;3;3",
"contribution_avg": 2.2,
"presentation": "2;2;3;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.155313"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xayT1nn8Mg | 2410.02847v1 | Deep Signature: Characterization of Large-Scale Molecular Dynamics | {
"content": "## Abstract\n\nAbstract Understanding protein dynamics are essential for deciphering protein functional mechanisms and developing molecular therapies. However, the complex high-dimensional dynamics and interatomic interactions of biological processes pose significant challenge for existing computational... | [
{
"id": "aBsVAopGA0",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces Deep Signature, a framework for characterizing dynamics on graphs. Deep Signature involves three key components: (1) graph coarsening using a... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "3;2;3",
"confidence_avg": 2.6666666666666665,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;3;2",
"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": "2026-01-14T22:16:04.156091"
} | {
"id": "rXjwCyI8Iv",
"metareview": "This paper introduces Deep Signature, a novel and computationally efficient framework for characterizing complex dynamics and interatomic interactions based on their evolving trajectories. The authors propose an approach that combines soft spectral clustering for local aggregati... | {
"decision": "Accept (Poster)"
} |
xdGsiYNfje | 2410.16638v2 | LLMScan: Causal Scan for LLM Misbehavior Detection | {
"content": "## Abstract\n\nAbstract Despite the success of Large Language Models (LLMs) across various fields, their potential to generate untruthful, biased and harmful responses poses significant risks, particularly in critical applications. This highlights the urgent need for systematic methods to detect and pre... | [
{
"id": "9KULVTY5xM",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces LLMSCAN, a novel method for detecting various types of misbehavior in Large Language Models (LLMs) through causal analysis of the models' in... | {
"rating": "3;3;5",
"rating_avg": 3.6666666666666665,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "1;2;3",
"soundness_avg": 2,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "1;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.156839"
} | {
"id": "xlBgEJbUI8",
"metareview": "After reading the reviewers' comments and reviewing the paper, we regret to recommend rejection. \n\nThe paper introduces a scan for detecting LLM misbehaviour called LLMScan, based on causality analysis of the model’s inner workings. While the idea is interesting, there are sev... | {
"decision": "Reject"
} |
xgQfWbV6Ey | 2407.08223v1 | Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting | {
"content": "## Abstract\n\nAbstract Retrieval augmented generation (RAG) combines the generative abilities of large language models (LLMs) with external knowledge sources to provide more accurate and up-to-date responses. Recent RAG advancements focus on improving retrieval outcomes through iterative LLM refinement... | [
{
"id": "9H94D6eaQI",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposed a retrieval-augmented generation framework termed SpeculativeRAG, leveraging high-level concepts analogical to speculative decoding.\nThe frame... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"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": "2026-01-14T22:16:04.157553"
} | {
"id": "sZhxkjfm9A",
"metareview": "The paper proposed Speculative RAG, a framework that decomposes retrieval-augmented generation into two stages. First, a specialist draft model produces multiple candidate responses from distinct document subsets. Then, a larger generalist LM verifies these drafts to select a fi... | {
"decision": "Accept (Poster)"
} |
xgtXkyqw1f | 2407.20183v1 | MindSearch: Mimicking Human Minds Elicits Deep AI Searcher | {
"content": "## Abstract\n\nAbstract Information seeking and integration is a complex cognitive task that consumes enormous time and effort.\nSearch engines reshape the way of seeking information but often fail to align with complex human intentions.\nInspired by the remarkable progress of Large Language Models (LLM... | [
{
"id": "EbGJQXVrsy",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes MindSearch, an LLM-based multi-agent information-seeking framework for complex multi-step information-seeking questions. MindSearch includes a ... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;4;3",
"contribution_avg": 3,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.158164"
} | {
"id": "7n2MSZnCEQ",
"metareview": "The paper proposes a novel framework that combines large language models (LLMs) with a multi-agent system for complex information-seeking tasks. The authors introduce two main components: WebPlanner, which decomposes user queries into a graph of sub-questions, and WebSearcher, w... | {
"decision": "Accept (Poster)"
} |
xiDJaTim3P | 2410.10114v2 | Mixture of Experts Made Personalized: Federated Prompt Learning for Vision-Language Models | {
"content": "## Abstract\n\nAbstract Prompt learning for pre-trained Vision-Language Models (VLMs) like CLIP has demonstrated potent applicability across diverse downstream tasks.\nThis lightweight approach has quickly gained traction from federated learning (FL) researchers who seek to efficiently adapt VLMs to het... | [
{
"id": "kgZdXySLA2",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces pFedMoAP that enables effective federated prompt learning for vision-language models like CLIP. The key innovation is allowing clients to do... | {
"rating": "3;6;6;6",
"rating_avg": 5.25,
"confidence": "4;3;5;2",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "1;2;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.158761"
} | {
"id": "LvnDyh4yex",
"metareview": "This paper introduces a lightweight federated prompt learning framework that uses a mixture of experts to personalize prompts and an attention-based gating network for efficient knowledge sharing across clients. It outperforms existing methods and allows direct updates of the pr... | {
"decision": "Accept (Poster)"
} |
xing7dDGh3 | 2410.05629v1 | Vector-ICL: In-context Learning with Continuous Vector Representations | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have shown remarkable in-context learning (ICL) capabilities on textual data.\nWe explore whether these capabilities can be extended to continuous vectors from diverse domains, obtained from black-box pretrained encoders.\nBy aligning input data with ... | [
{
"id": "ApGlGSj7Gp",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper studies the feasibility of vector-ICL, that extends the in-context learning capabilities of LLMs to continuous vectors. Authors use light-weight project... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;2;4;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;4;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.159477"
} | {
"id": "xSCSz9to3J",
"metareview": "This paper proposes Vector-ICL which encodes input demonstrations into continuous vectors via pretrained projectors and fine-tuning process to elicit LLMs' capability of understanding both vector and textual inputs. This strategy enables effective use of the ICL framework in dom... | {
"decision": "Accept (Poster)"
} |
xiyzCfXTS6 | 2409.18582v1 | Optimistic Games for Combinatorial Bayesian Optimization with Application to Protein Design | {
"content": "## Abstract\n\nAbstract Bayesian optimization (BO) is a powerful framework to optimize black-box expensive-to-evaluate functions via sequential interactions.\nIn several important problems (e.g. drug discovery, circuit design, neural architecture search, etc.), though, such functions are defined over la... | [
{
"id": "J59lpu12a9",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes a novel method that frames the protein design task as an optimization problem of finding a Nash Equilibrium (NE) with unknown utility functions... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "3;3;3;4",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;4",
"contribution_avg": 2.5,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.160405"
} | {
"id": "raUJ5ysXVZ",
"metareview": "The paper introduces GameOpt, a game-theoretical framework for combinatorial Bayesian optimization (CBO), with a focus on protein design. This work represents an innovative contribution by leveraging cooperative game theory to tackle scalability challenges in high-dimensional di... | {
"decision": "Accept (Poster)"
} |
xizpnYNvQq | 2410.04468v1 | Revisiting In-context Learning Inference Circuit in Large Language Models | {
"content": "## Abstract\n\nAbstract I n- c ontext L earning (ICL) is an emerging few-shot learning paradigm on L anguage M odels (LMs) with inner mechanisms un-explored. There are already existing works describing the inner processing of ICL, while they struggle to capture all the inference phenomena in large langu... | [
{
"id": "fzb5M5P4xB",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper aims to explain the mechanisms behind in-context learning (ICL) using the inference circuit framework.\n\nAccording to the authors, the ICL process con... | {
"rating": "6;6;6;8",
"rating_avg": 6.5,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;2;2;3",
"contribution_avg": 2.5,
"presentation": "4;2;2;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.161416"
} | {
"id": "5ESZrB0oOF",
"metareview": "**Summary:**\nThe paper investigates the mechanisms behind In-Context Learning in large language models. The authors introduce a three-step inference circuit comprising (1) input text encoding, (2) semantic merge, and (3) feature retrieval and copying. Through empirical analysis... | {
"decision": "Accept (Poster)"
} |
xlbXRJ2XCP | 2409.05100v2 | MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks | {
"content": "## Abstract\n\nAbstract We propose a novel approach to compute the MAXCUT in attributed graphs, i.e. , graphs with features associated with nodes and edges. Our approach is robust to the underlying graph topology and is fully differentiable, making it possible to find solutions that jointly optimize the... | [
{
"id": "OyueBNpX76",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces an innovative method for computing the MAXCUT in attributed graphs which is fully differentiable, enabling joint optimization of the MAXCUT ... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;2;3;4",
"confidence_avg": 3.25,
"soundness": "3;3;3;2",
"soundness_avg": 2.75,
"contribution": "1;3;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": "2026-01-14T22:16:04.162296"
} | {
"id": "WMMpQGpoAy",
"metareview": "**(a) Scientific Claims and Findings:**\nThe paper introduces a novel approach for computing the MAXCUT in attributed graphs—graphs with features associated with nodes and edges. This method is fully differentiable, enabling joint optimization of the MAXCUT alongside other objec... | {
"decision": "Accept (Poster)"
} |
xlrpVyMIwz | 2409.18865v1 | Positional Encoder Graph Quantile Neural Networks for Geographic Data | {
"content": "## Abstract\n\nAbstract Positional Encoder Graph Neural Networks (PE-GNNs) are a leading approach for modeling continuous spatial data. However, they often fail to produce calibrated predictive distributions, limiting their effectiveness for uncertainty quantification.\nWe introduce the Positional Encod... | [
{
"id": "zDrwrsGFTQ",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents the Positional Encoder Graph Quantile Neural Network (PE-GQNN), a novel model for spatial data prediction. PE-GQNN integrates Positional Encod... | {
"rating": "1;3;5;6",
"rating_avg": 3.75,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;1;2;3",
"contribution_avg": 2,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.163048"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xnF2U0ro7b | 2405.18183v1 | Feature-Based Online Bilateral Trade | {
"content": "## Abstract\n\nAbstract Bilateral trade models the problem of facilitating trades between a seller and a buyer having private valuations for the item being sold.\nIn the online version of the problem, the learner faces a new seller and buyer at each time step, and has to post a price for each of the two... | [
{
"id": "t4JTPKzpwk",
"initial_rating": 8,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 2,
"summary": "This paper tackles the bilateral trade problem in an online setting where there is an additional context present. At each time step $t$, a buyer and a seller arri... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;4;4",
"soundness_avg": 3.25,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;2",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.163979"
} | {
"id": "YdbvK7yKKN",
"metareview": "The reviewers all appreciate the new scouting algorithm and the minimax optimal regret it achieves. A good paper overall.",
"additional_comments": "NA"
} | {
"decision": "Accept (Poster)"
} |
xof0bvftR1 | 2405.20448v2 | Knockout: A simple way to handle missing inputs | {
"content": "## Abstract\n\nAbstract Deep learning models can extract predictive and actionable information from complex inputs.\nThe richer the inputs, the better these models usually perform.\nHowever, models that leverage rich inputs (e.g., multi-modality) can be difficult to deploy widely, because some inputs ma... | [
{
"id": "Bj5UMUPAAd",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces \"Knockout,\" a simple yet effective data augmentation strategy for handling missing inputs during inference in machine learning models. Knoc... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.165384"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xrXci5YGm7 | 2410.07041v1 | Emergent properties with repeated examples | {
"content": "## Abstract\n\nAbstract We study the performance of transformers as a function of the number of repetitions of training examples with algorithmically generated datasets. On three problems of mathematics: the greatest common divisor, modular multiplication, and matrix eigenvalues, we show that for a fixe... | [
{
"id": "8MpQ4B9XiG",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper empirically shows that the repetition of training data can be beneficial in certain setups. The authors conduct experiments on three algorithmically ge... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;3;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.166134"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xreOs2yjqf | 2406.16562v3 | EvalAlign: Supervised Fine-Tuning Multimodal LLMs with Human-Aligned Data for Evaluating Text-to-Image Models | {
"content": "## Abstract\n\nAbstract The recent advancements in text-to-image generative models have been remarkable. Yet, the field suffers from a lack of evaluation metrics that accurately reflect the performance of these models, particularly lacking fine-grained metrics that can guide the optimization of the mode... | [
{
"id": "KHkqh06Th8",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes EvalAlign, a metric characterized by accuracy, stability, and fine-grainedness. Evaluation on 24 text-to-image generation models shows that Ev... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;4;5;4",
"confidence_avg": 4.5,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;2;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.166892"
} | {
"id": "pRlIvbiTi8",
"metareview": "This paper collects a human-annotated preference dataset to fine-tune MLLMs as evaluators for T2I models.\n\nThe strength of the paper is: 1) a fine-grained evaluation data sets 2) Experiments on large number (24) generative models \n\nHowever, the major weaknesses are: 1) the e... | {
"decision": "Reject"
} |
xriJVaTh4C | 2403.07095v3 | Gaussian Loss Smoothing Enables Certified Training with Tight Convex Relaxations | {
"content": "## Abstract\n\nAbstract Training neural networks with high certified accuracy against adversarial examples remains an open problem despite significant efforts. While certification methods can effectively leverage tight convex relaxations for bound computation, in training, these methods perform worse th... | [
{
"id": "5YKHx42qTR",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes to use a method (Gaussian Loss Smoothing) to address the paradox of certified training which is caused by the discontinuity/non-smoothness/sen... | {
"rating": "1;3;6",
"rating_avg": 3.3333333333333335,
"confidence": "4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "1;1;3",
"contribution_avg": 1.6666666666666667,
"presentation": "1;3;3",
"presentation_avg": 2.3333333333333... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.167620"
} | {
"id": "056NmBlz9p",
"metareview": "The main claim of the paper is that issues arising in certified training can be alleviated by smoothing the loss surface in parameter space of a neural network by convolving it with a Gaussian distribution. \n\nCombining ideas of loss smoothing and convex relaxation is an intere... | {
"decision": "Reject"
} |
xsELpEPn4A | 2310.17631v1 | JudgeLM: Fine-tuned Large Language Models are Scalable Judges | {
"content": "## Abstract\n\nAbstract Evaluating Large Language Models (LLMs) in open-ended scenarios is challenging because existing benchmarks and metrics can not measure them comprehensively.\nTo address this problem, we propose to fine-tune LLMs as scalable judges (JudgeLM) to evaluate LLMs efficiently and effect... | [
{
"id": "Kow3ZtQ7cL",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This paper targets evaluating and building LLM's specifically for judging answer correctness on open-ended tasks. To do this, they construct a dataset which consi... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.168222"
} | {
"id": "RpoJjFr1qB",
"metareview": "This paper introduces resources to enable training of JudgeLM, a language model trained specifically to behave as an automatic evaluator. A large-scale dataset to train judge models is proposed, covering diverse seed tasks. The data also includes generated answers and detailed r... | {
"decision": "Accept (Spotlight)"
} |
xuQSp75HmP | 2409.15278v2 | PixWizard: Versatile Image-to-Image Visual Assistant with Open-Language Instructions | {
"content": "## Abstract\n\nAbstract This paper presents a versatile image-to-image visual assistant, PixWizard , designed for image generation, manipulation, and translation based on free-from language instructions. To this end, we tackle a variety of vision tasks into a unified image-text-to-image generation frame... | [
{
"id": "0Shh3VP1Ov",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces PixWizard, a visual assistant that performs diverse image generation, manipulation, and translation tasks based on free-form language instru... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.169006"
} | {
"id": "yto0EzM6zH",
"metareview": "All the 4 reviewers provide positive ratings after rebuttal, with 2 upgraded score. Initially, the reviewers had concerns about technical contributions, performance gains compared to task-specific models and prior work like Emu Edit, and generalization ability to various instruc... | {
"decision": "Accept (Poster)"
} |
xvUVk9T3kZ | 2402.11367v1 | Multi Task Inverse Reinforcement Learning for Common Sense Reward | {
"content": "## Abstract\n\nAbstract One of the challenges in applying reinforcement learning in a complex real-world environment lies in providing the agent with a sufficiently detailed reward function. Any misalignment between the reward and the desired behavior can result in unwanted outcomes. This may lead to is... | [
{
"id": "8NX9GH6s79",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "This work proposes a method to recover a “common-sense” reward using demonstrations across multiple tasks. This reward represents the part of the reward function ... | {
"rating": "1;1;5;5",
"rating_avg": 3,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "1;1;2;2",
"soundness_avg": 1.5,
"contribution": "2;1;2;2",
"contribution_avg": 1.75,
"presentation": "3;3;3;1",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.169788"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xy6B5Fh2v7 | 2410.07176v1 | Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models | {
"content": "## Abstract\n\nAbstract Retrieval-Augmented Generation (RAG), while effective in integrating external knowledge to address the limitations of large language models (LLMs), can be undermined by imperfect retrieval, which may introduce irrelevant, misleading, or even malicious information. Despite its imp... | [
{
"id": "YOhj8eHKcw",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper introduces Astute RAG, a novel approach addressing imperfect retrieval and knowledge conflicts in RAG systems. \nThe authors first conduct comprehensive... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.171169"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
xzKFnsJIXL | 2405.14457v2 | Tighter Privacy Auditing of DP-SGD in the Hidden State Threat Model | {
"content": "## Abstract\n\nAbstract Machine learning models can be trained with formal privacy guarantees via differentially private optimizers such as DP-SGD. In this work, we focus on a threat model where the adversary has access only to the final model, with no visibility into intermediate updates. In the litera... | [
{
"id": "lXalOk7mcN",
"initial_rating": 5,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "In the \"hidden state\" threat model for DP-SGD, an adversary does not have access to intermediate updates and can see only the final model. This paper proposes t... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "3;2;3;4",
"confidence_avg": 3,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"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": "2026-01-14T22:16:04.171889"
} | {
"id": "zo6D601ptO",
"metareview": "The paper examines the extent to which not revealing the intermediate states provides additional privacy for training ML models. Prior work has shown \"privacy amplification via iteration\" when intermediate states are not revealed, for convex objectives. The paper shows that wh... | {
"decision": "Accept (Poster)"
} |
xzSUdw6s76 | 2410.05315v1 | PALMBENCH: A COMPREHENSIVE BENCHMARK OF COMPRESSED LARGE LANGUAGE MODELS ON MOBILE PLATFORMS | {
"content": "## \\thesubsection Multi-Turn Questions\n\nMT-bench [MTBenchArena_2023] employs a predefined multi-turn question set to assess conversational models across turns for instruction following and dialog coherence, critical capabilities for human-like interaction. Each sequence begins with an initial questio... | [
{
"id": "Q5QBMrzfi0",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper describes the benchmarking results of several quantized Large Language Models on smartphones and edge devices. Specifically, it quantifies the CPU, GPU... | {
"rating": "5;5;5;6;8",
"rating_avg": 5.8,
"confidence": "3;4;5;4;4",
"confidence_avg": 4,
"soundness": "2;3;2;3;4",
"soundness_avg": 2.8,
"contribution": "2;2;2;2;3",
"contribution_avg": 2.2,
"presentation": "3;3;3;3;4",
"presentation_avg": 3.2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.172505"
} | {
"id": "0Eb17mf0Iu",
"metareview": "The paper presents PalmBench, a benchmarking framework designed to assess the performance of compressed Large Language Models (LLMs) on mobile devices. It offers a comprehensive evaluation of LLMs across various quantization configurations and mobile platforms with diverse hardw... | {
"decision": "Accept (Poster)"
} |
y10AP0BkID | 2408.15708v1 | Towards Realistic Example-based Modeling via 3D Gaussian Stitching | {
"content": "## Abstract\n\nAbstract. Using parts of existing models to rebuild new models, commonly termed as example-based modeling, is a classical methodology in the realm of computer graphics. Previous works mostly focus on shape composition, making them very hard to use for realistic composition of 3D objects c... | [
{
"id": "AkdH8ejShI",
"initial_rating": 3,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes a method for interactively editing radiance fields encoded in point clouds (\"Guassian Splatting\"). It consists of two components: First, poin... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;2;4;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;2;2;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:04.172977"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
y15LAM4u0A | 2410.09604v1 | EmbodiedCity: A Benchmark Platform for Embodied Agent in Real-world City Environment | {
"content": "## Abstract\n\nAbstract Embodied artificial intelligence (EmbodiedAI) emphasizes the role of an agent’s body in generating human-like behaviors. The recent efforts on EmbodiedAI pay a lot of attention to building up machine learning models to possess perceiving, planning, and acting abilities, thereby e... | [
{
"id": "LOXscdXOWg",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents a comprehensive benchmark platform aimed at assessing the performance of embodied agents in a realistic urban setting. Unlike previous benchma... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "4;5;4;3",
"confidence_avg": 4,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.173650"
} | {
"id": "LztXHnntu5",
"metareview": "This paper was reviewed by four field experts and received unanimously negative evaluations. The main concerns raised include a lack of significant technical contributions and relatively underwhelming results. Additionally, no rebuttal was provided by the authors. The AC finds n... | {
"decision": "Reject"
} |
y1iU5czYpE | 2408.15664v1 | Auxiliary-Loss-Free Load Balancing Strategy for Mixture-of-Experts | {
"content": "## Abstract\n\nAbstract For Mixture-of-Experts (MoE) models, an unbalanced expert load will lead to routing collapse or increased computational overhead.\nExisting methods commonly employ an auxiliary loss to encourage load balance, but a large auxiliary loss will introduce non-negligible interference g... | [
{
"id": "AaMezXg6gn",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The authors propose an alternative loss-free method for load balancing of experts during MoE training. Load imbalance is a critical issue in MoE training as it ca... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;4;4;2",
"confidence_avg": 3.5,
"soundness": "1;2;2;3",
"soundness_avg": 2,
"contribution": "1;2;2;2",
"contribution_avg": 1.75,
"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": "2026-01-14T22:16:04.174240"
} | {
"id": "oyjk5ozJEL",
"metareview": "After reviewing the authors' rebuttal and considering the discussion phase, all reviewers maintained reservations about this submission, resulting in final ratings of 3 (Reviewer tLSf), 5 (Reviewer TZLQ), 5 (Reviewer 8VSt), and 3 (Reviewer QrX9). \n\nWhile the rebuttal partially... | {
"decision": "Reject"
} |
y3CdSwREZl | 2410.04819v1 | MINER: Mining the Underlying Pattern of Modality-Specific Neurons in Multimodal Large Language Models | {
"content": "## Abstract\n\nAbstract In recent years, multimodal large language models (MLLMs) have significantly advanced, integrating more modalities into diverse applications. However, the lack of explainability remains a major barrier to their use in scenarios requiring decision transparency. Current neuron-leve... | [
{
"id": "EgMEgZuqa5",
"initial_rating": 3,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 1,
"summary": "This paper proposes a framework named MINER to understand the underlying pattern of modality-specific neurons in multimodal large language models (MLLM). The fram... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "4;3;4;3;3",
"confidence_avg": 3.4,
"soundness": "3;3;2;3;3",
"soundness_avg": 2.8,
"contribution": "2;3;3;2;3",
"contribution_avg": 2.6,
"presentation": "1;3;3;2;3",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:04.174870"
} | {
"id": "3c4A92o0mA",
"metareview": "The paper proposes MINER, a framework for identifying modality-specific neurons. The paper is well-motivated and has an interesting concept with a novel direction. However, key shortcomings include unclear design choices for the experiments as well as unanswered questions regard... | {
"decision": "Reject"
} |
y3zswp3gek | 2410.01524v2 | HarmAug: Effective Data Augmentation for Knowledge Distillation of Safety Guard Models | {
"content": "## Abstract\n\nAbstract Safety guard models that detect malicious queries aimed at large language models (LLMs) are essential for ensuring the secure and responsible deployment of LLMs in real-world applications.\nHowever, deploying existing safety guard models with billions of parameters alongside LLMs... | [
{
"id": "Td8iy0762W",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "Safety guard models (SGMs) are often deployed along with a target large language model in order to detect malicious and unsafe queries. They are also used for red... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "2;4;4;4",
"confidence_avg": 3.5,
"soundness": "3;4;4;4",
"soundness_avg": 3.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "4;4;3;4",
"presentation_avg": 3.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.175658"
} | {
"id": "jo9EweIJDz",
"metareview": "This paper proposes HarmAug, a data augmentation method to create efficient safety guard models for LLMs through knowledge distillation. The key finding is that a 435M parameter model trained with HarmAug achieves comparable or better performance than 7B+ parameter models on saf... | {
"decision": "Accept (Poster)"
} |
y4DtzADzd1 | 2411.04873v1 | Boosting Latent Diffusion with Perceptual Objectives | {
"content": "## Abstract\n\nAbstract Latent diffusion models (LDMs) power state-of-the-art high-resolution generative image models. LDMs learn the data distribution in the latent space of an autoencoder (AE) and produce images by mapping the generated latents into RGB image space using the AE decoder. While this app... | [
{
"id": "0Y5ujrbwMX",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper studies the perceptual loss function in the training of diffusion models and proposes to compare the features between $z_0$ and $\\hat{z}_0$ by sending ... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:04.176789"
} | {
"id": "So9cYEDfzZ",
"metareview": "Summary: Proposed a perceptual loss in the context of latent diffusion models to improve image sharpness. It uses features extracted from the latent decoder to calculate the distance between z0 and the predicted z0_hat at the pyramid levels of the decoder. Demonstrates improved ... | {
"decision": "Accept (Poster)"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.