paper_id string | arxiv_id string | title string | markdown dict | reviews list | scores dict | metadata dict | meta_review dict | decision dict |
|---|---|---|---|---|---|---|---|---|
6BoStmXGBf | 2406.10973v2 | ExPLoRA: Parameter-Efficient Extended Pre-Training to Adapt Vision Transformers under Domain Shifts | {
"content": "## Abstract\n\nAbstract Parameter-efficient fine-tuning (PEFT) techniques such as low-rank adaptation (LoRA) can effectively adapt large pre-trained foundation models to downstream tasks using only a small fraction (0.1%-10%) of the original trainable weights.\nAn under-explored question of PEFT is in e... | [
{
"id": "z5F51RDUms",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This work presents ExPLoRA, which initializes a ViT with pre-trained weights, selectively unfreezes 1 - 2 blocks, tunes remaining weights with LoRA, and continues... | {
"rating": "5;5;5;5;6",
"rating_avg": 5.2,
"confidence": "5;3;4;5;5",
"confidence_avg": 4.4,
"soundness": "3;3;3;2;4",
"soundness_avg": 3,
"contribution": "2;2;2;2;4",
"contribution_avg": 2.4,
"presentation": "4;3;3;3;2",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.894089"
} | {
"id": "Su1UjRxF9B",
"metareview": "This work introduces ExPLoRA, a method that initializes a Vision Transformer (ViT) with pre-trained weights, selectively unfreezes one to two blocks, fine-tunes the remaining weights using LoRA, and continues unsupervised pre-training on a new domain. Subsequently, the model und... | {
"decision": "Reject"
} |
6ESRicalFE | 2410.11143v1 | LLM Unlearning via Loss Adjustment with Only Forget Data | {
"content": "## Abstract\n\nAbstract Unlearning in Large Language Models (LLMs) is essential for ensuring ethical and responsible AI use, especially in addressing privacy leak, bias, safety, and evolving regulations. Existing approaches to LLM unlearning often rely on retain data or a reference LLM, yet they struggl... | [
{
"id": "ZoyRyWwrW6",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces FLAT (Forget data only Loss Adjustment), a novel method for unlearning in large language models (LLMs) without requiring retain data or a ref... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "3;2;4;3",
"confidence_avg": 3,
"soundness": "2;2;4;3",
"soundness_avg": 2.75,
"contribution": "2;2;4;3",
"contribution_avg": 2.75,
"presentation": "1;2;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.895059"
} | {
"id": "GJUs2h6pvv",
"metareview": "This paper proposes a new approach to unlearning. The idea is to only use the desired forget set (rather than a retain set or any auxiliary model, such as a reference model), and to use a particular form of loss function adjustment to perform unlearning. The loss function adjust... | {
"decision": "Accept (Poster)"
} |
6EUtjXAvmj | 2410.09945v1 | Variational Diffusion Posterior Sampling with Midpoint Guidance | {
"content": "## Abstract\n\nAbstract Diffusion models have recently shown considerable potential in solving Bayesian inverse problems when used as priors. However, sampling from the resulting\ndenoising posterior distributions remains a challenge as it involves intractable terms. To tackle this issue, state-of-the-a... | [
{
"id": "OAK1izDebt",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a novel diffusion-based posterior sampling method called Midpoint Guidance Posterior Sampling (MGPS) to address Bayesian inverse problems. I... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "4;4;5;3",
"confidence_avg": 4,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "3;4;3;3",
"contribution_avg": 3.25,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.896045"
} | {
"id": "J5scty68x0",
"metareview": "The paper proposes a diffusion-based method for posterior sampling in diffusion models. The four reviewers all indicated that the paper is clearly above threshold for acceptance. They found the paper to be written well and present an idea that could be found widely useful. The e... | {
"decision": "Accept (Oral)"
} |
6EkWIfvjj9 | 2410.20088v1 | RARe: Retrieval Augmented Retrieval with In-Context Examples | {
"content": "## Abstract\n\nAbstract We investigate whether in-context examples, widely used in decoder-only language models (LLMs), can improve embedding model performance in retrieval tasks. Unlike in LLMs, naively prepending in-context examples (query-document pairs) to the target query at inference time does not... | [
{
"id": "dcTG27KQAO",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper explores the application of in-context learning to improve the performance of retriever models in information retrieval tasks. The authors introduce RAR... | {
"rating": "5;5;5;5",
"rating_avg": 5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "3;3;2;2",
"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": "2026-01-14T22:16:00.897025"
} | {
"id": "DQbfgK0OCa",
"metareview": "This paper proposes a novel framework, RARe, which augments the input with in-context learning examples for retrieving relevant documents. Different from existing ICL which helps LLMs during the inference/generation process, RARe explores the benefit of ICL in producing semantic... | {
"decision": "Reject"
} |
6GWvBa60LZ | 2409.17872v1 | A method for identifying causality in the response of nonlinear dynamical systems | {
"content": "## Abstract\n\nAbstract Predicting the response of nonlinear dynamical systems subject to random, broadband excitation is important across a range of scientific disciplines, such as structural dynamics and neuroscience. Building data-driven models requires experimental measurements of the system input a... | [
{
"id": "OrFudiBOt6",
"initial_rating": 3,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 1,
"summary": "The paper introduces a method for estimating the coherence $Co(y, y_n)$ between the observed noisy data $y_n$, and the “true” underlying system $y$. This estimate... | {
"rating": "3;3;3",
"rating_avg": 3,
"confidence": "2;2;3",
"confidence_avg": 2.3333333333333335,
"soundness": "2;2;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;2",
"contribution_avg": 2,
"presentation": "1;2;1",
"presentation_avg": 1.3333333333333333
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.897937"
} | {
"id": "LfrpfnzMeA",
"metareview": "The authors aim in this work to provide a measure by which to judge the best possible dynamical system model that can fit experimentally measured inputs and outputs of a system. The measure focuses on a frequency based approximation of coherence, and applies the metric to a num... | {
"decision": "Reject"
} |
6H4jRWKFc3 | 2312.08598v1 | MotherNet: Fast Training and Inference via Hyper-Network Transformers | {
"content": "## Abstract\n\nAbstract The advent of Foundation Models is transforming machine learning across many modalities (e.g., language, images, videos) with prompt engineering replacing training in many settings. Recent work on tabular data (e.g., TabPFN ) hints at a similar opportunity to build Foundation Mod... | [
{
"id": "qln8G23j4b",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "The authors propose a novel transformer based hypernetwork model that can be 'in-context' prompted with a supervised dataset and it can generate weights for a sma... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "2;2;2;4",
"soundness_avg": 2.5,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "1;4;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.898592"
} | {
"id": "BQLbHrV5mh",
"metareview": "The paper introduces a hypernetwork for in-context learning on tabular datasets, inspired by TabPFN and meta-learning paradigms. MotherNet generates task-specific models via a pre-trained transformer, allowing for inference without gradient descent or hyperparameter tuning. The ... | {
"decision": "Accept (Poster)"
} |
6HcnC3pPkp | 2407.12863v1 | Token-Supervised Value Models for Enhancing Mathematical Problem-Solving Capabilities of Large Language Models | {
"content": "## Abstract\n\nAbstract Large Language Models (LLMs) have demonstrated impressive problem-solving capabilities in mathematics through step-by-step reasoning chains.\nHowever, they are susceptible to reasoning errors that impact the quality of subsequent reasoning chains and the final answer due to langu... | [
{
"id": "XNBDOY5LZL",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes token-supervised value models (TVMs) as a superior way to verify whether tree-search based math reasoning models are on the right track or not... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;4;3",
"contribution_avg": 3,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.899217"
} | {
"id": "7SsUf3l4bq",
"metareview": "This work introduces token-supervised value models (TVMs), which provide token-level supervision using verifiers. The verifiers assign a score to each token, indicating the probability of reaching a correct final answer. The authors show that TVMs achieve lower false negative er... | {
"decision": "Accept (Poster)"
} |
6Imw3BwOMo | 2306.11128v2 | CAMMARL: Conformal Action Modeling in Multi Agent Reinforcement Learning | {
"content": "## Abstract\n\nAbstract Before taking actions in an environment with more than one intelligent agent, an autonomous agent may benefit from reasoning about the other agents and utilizing a notion of a guarantee or confidence about the behavior of the system. In this article, we propose a novel multi-agen... | [
{
"id": "uwkLxN4SfV",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper proposes to extend MARL by providing agents with conformal predictions of other agents’ actions. By numerical experiments, the authors demonstrate convi... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;4;3;3",
"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;2;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.899958"
} | {
"id": "FNF2i08Jpf",
"metareview": "This paper proposes CAMMARL, a new algorithm for multi-agent reinforcement learning that uses conformal prediction to model the actions of other agents as probabilistic sets, i.e. conformal action sets, to improve cooperative decision-making under uncertainty. The proposed frame... | {
"decision": "Reject"
} |
6Ire5JaobL | 2410.03229v1 | Elucidating the Design Choice of Probability Paths in Flow Matching for Forecasting | {
"content": "## Abstract\n\nAbstract Flow matching has recently emerged as a powerful paradigm for generative modeling and has been extended to probabilistic time series forecasting in latent spaces. However, the impact of the specific choice of probability path model on forecasting performance remains under-explore... | [
{
"id": "G5TFfTHSVJ",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper addresses spatio-temporal forecasting using latent flow matching. In spatio-temporal forecasting, the performance of flow matching is highly dependent ... | {
"rating": "5;5;6",
"rating_avg": 5.333333333333333,
"confidence": "3;3;3",
"confidence_avg": 3,
"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": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.900719"
} | {
"id": "KZL150J7Wo",
"metareview": "The paper introduces a new probability path model under the framework of latent flow matching to improve forecasting performance for spatio-temporal data. The novelty resides in implementing a new probability path for flow matching within the latent space of a pre-trained autoen... | {
"decision": "Reject"
} |
6JDpWJrjyK | 2406.19705v5 | DISCO: Efficient Diffusion Solver for Large-Scale Combinatorial Optimization Problems | {
"content": "## Abstract\n\nAbstract Combinatorial Optimization (CO) problems are fundamentally important in numerous real-world applications across diverse industries, characterized by entailing enormous solution space and demanding time-sensitive response.\nDespite recent advancements in neural solvers, their limi... | [
{
"id": "SJs3F0hQan",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces DISCO, a diffusion-based solver optimized for large-scale combinatorial optimization (CO) problems, such as the Traveling Salesman Problem (T... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.901574"
} | {
"id": "lfQZ9Oj5dG",
"metareview": "(a) Summarize the scientific claims and findings: The paper claims DISCO achieves faster inference times and higher accuracy on large-scale combinatorial optimization problems like TSP and MIS by introducing residue-constrained denoising and a multi-modal graph search, but revie... | {
"decision": "Reject"
} |
6KZ80APcxf | 2411.02470v1 | Benchmarking XAI Explanations with Human-Aligned Evaluations | {
"content": "## Abstract\n\nAbstract In this paper, we introduce PASTA (Perceptual Assessment System for explanaTion of Artificial intelligence), a novel framework for a human-centric evaluation of XAI techniques in computer vision.\nOur first key contribution is a human evaluation of XAI explanations on four divers... | [
{
"id": "XoTbCTYZT5",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces PASTA, a perceptual assessment system designed to benchmark explainable AI (XAI) techniques in a human-centric manner. The authors first inte... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "2;4;4;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "1;3;2;3",
"contribution_avg": 2.25,
"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": "2026-01-14T22:16:00.902401"
} | {
"id": "jYutUK9rUu",
"metareview": "The submission initially had mixed reviews, and the major concerns raised are:\n\n1. limited benchmark size (5 humans, 100 images) [cj8g, 2wcf]\n2. share images in training/test sets? [cj8g]\n3. missing ablation study on grounding method [cj8g]\n4. missing explanation/analysis a... | {
"decision": "Reject"
} |
6LKmaC4cO0 | 2410.11001v1 | Graph of Records: Boosting Retrieval Augmented Generation for Long-context Summarization with Graphs | {
"content": "## Abstract\n\nAbstract Retrieval-augmented generation (RAG) has revitalized Large Language Models (LLMs) by injecting non-parametric factual knowledge.\nCompared with long-context LLMs, RAG is considered an effective summarization tool in a more concise and lightweight manner, which can interact with L... | [
{
"id": "gvfIPLoQu9",
"initial_rating": 6,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces the Graph of Records (GoR), a novel method designed to enhance retrieval-augmented generation (RAG) systems for long-context summarization us... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "3;4;4;3",
"confidence_avg": 3.5,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "2;1;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.903290"
} | {
"id": "rg8aKsMONJ",
"metareview": "**Summary:** \nThe paper proposes Graph of Records (GoR), a method to enhance RAG for long-context summarization by leveraging LLM-generated historical responses, which are often neglected. GoR constructs a graph that links retrieved text chunks with their corresponding LLM resp... | {
"decision": "Reject"
} |
6LtdZCyuZR | 2407.12843v4 | NutriBench: A Dataset for Evaluating Large Language Models in Nutrition Estimation from Meal Descriptions | {
"content": "## Abstract\n\nAbstract Accurate nutrition estimation helps people make informed dietary choices and is essential in the prevention of serious health complications. We present NutriBench , the first publicly available natural language meal description nutrition benchmark. NutriBench consists of 11,857 m... | [
{
"id": "CJ3lfsniGv",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents NUTRIBENCH, a new benchmark dataset designed to evaluate LLMs on nutrition estimation from natural language meal descriptions. The dataset inc... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;3;4;3",
"confidence_avg": 3.75,
"soundness": "3;3;2;3",
"soundness_avg": 2.75,
"contribution": "3;2;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.903924"
} | {
"id": "cF1OkOu21H",
"metareview": "This paper introduces NutriBench, a benchmark dataset with 11,857 natural language meal descriptions annotated with macronutrient data to evaluate LLMs for nutrition estimation. The authors conduct thorough experiments on 12 LLMs, explore prompting strategies, compare prediction... | {
"decision": "Accept (Poster)"
} |
6MBqQLp17E | 2410.03462v2 | Linear Transformer Topological Masking with Graph Random Features | {
"content": "## Abstract\n\nAbstract When training transformers on graph-structured data, incorporating information about the underlying topology is crucial for good performance. Topological masking , a type of relative position encoding, achieves this by upweighting or downweighting attention depending on the relat... | [
{
"id": "4VUgSfDxFS",
"initial_rating": 3,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper addresses the challenge of incorporating graph structural information into transformer attention mechanisms while maintaining computational efficiency.... | {
"rating": "3;6;6;6;6;8",
"rating_avg": 5.833333333333333,
"confidence": "5;2;2;3;2;3",
"confidence_avg": 2.8333333333333335,
"soundness": "3;3;3;3;3;3",
"soundness_avg": 3,
"contribution": "3;2;3;3;3;3",
"contribution_avg": 2.8333333333333335,
"presentation": "4;3;3;1;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.904756"
} | {
"id": "x21DNXb68o",
"metareview": "The authors consider the transformer attention with graph structured data. In particular, the authors focus on the topological masking of low-rank attention. The authors propose to leverage the graph random features to approximate topological masks and parameterize it as a learn... | {
"decision": "Accept (Poster)"
} |
6Mdvq0bPyG | 2407.11062v2 | EfficientQAT: Efficient Quantization-Aware Training for Large Language Models | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) are crucial in modern natural language processing and artificial intelligence. However, they face challenges in managing their significant memory requirements. Although quantization-aware training (QAT) offers a solution by reducing memory consumption... | [
{
"id": "XvHssWoXM8",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes EfficientQAT, a novel quantization-aware training (QAT) framework tailored for large language models (LLMs). Aiming to address the high memory ... | {
"rating": "3;3;3",
"rating_avg": 3,
"confidence": "4;5;4",
"confidence_avg": 4.333333333333333,
"soundness": "2;2;2",
"soundness_avg": 2,
"contribution": "1;2;2",
"contribution_avg": 1.6666666666666667,
"presentation": "2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.905663"
} | {
"id": "7gjyN0jxmy",
"metareview": "This paper introduces a quantization-aware training framework called EfficientQAT designed for large language models. The method is two-phase and combines block-wise training and end-to-end tuning of quantization parameters. While the idea is practical and shows promise in impro... | {
"decision": "Reject"
} |
6Mg7pjG7Sw | 2410.07610v1 | CSA: Data-efficient Mapping of Unimodal Features to Multimodal Features | {
"content": "## Abstract\n\nAbstract Multimodal encoders like CLIP excel in tasks such as zero-shot image classification and cross-modal retrieval. However, they require excessive training data.\nWe propose canonical similarity analysis (CSA), which uses two unimodal encoders to replicate multimodal encoders using l... | [
{
"id": "ue2ulqHQ1z",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes a new approach called Canonical Similarity Analysis (CSA) that addresses the challenge of training multimodal encoders, like CLIP, which typica... | {
"rating": "5;5;6;6;6",
"rating_avg": 5.6,
"confidence": "4;4;3;3;3",
"confidence_avg": 3.4,
"soundness": "3;3;3;3;3",
"soundness_avg": 3,
"contribution": "3;2;2;3;2",
"contribution_avg": 2.4,
"presentation": "2;2;3;4;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.906319"
} | {
"id": "XTcDsZkThr",
"metareview": "This paper aims to design an efficient scheme for fast multi-modal learning. With frozen CLIP or the like, they build their model with canonical similarity (deduced from CCA) to measure the similarity from unimodal to multimodal feature spaces. As the authors claimed, the main s... | {
"decision": "Accept (Poster)"
} |
6Mxhg9PtDE | 2406.05946v1 | Safety Alignment Should be Made More Than Just a Few Tokens Deep | {
"content": "## Abstract\n\nAbstract The safety alignment of current Large Language Models (LLMs) is vulnerable. Relatively simple attacks, or even benign fine-tuning, can jailbreak aligned models. We argue that many of these vulnerabilities are related to a shared underlying issue: safety alignment can take shortcu... | [
{
"id": "Vu2T2ozfzB",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "This paper proposes that the fragility of LLMs to various attacks (adversarial, prefilling, sampling, fine-tuning) could be explained by the models taking “shortc... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "5;5;2;4",
"confidence_avg": 4,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "3;2;4;3",
"contribution_avg": 3,
"presentation": "4;2;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Oral",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.907226"
} | {
"id": "1tX3yYpoxV",
"metareview": "This paper introduces the concept of \"shallow safety alignment\" to uncover a fundamental vulnerability in current safety alignment approaches and proposes \"deep safety alignment\" as a promising defense. All reviewers agreed that this work addresses a highly relevant and time... | {
"decision": "Accept (Oral)"
} |
6NNA0MxhCH | 2407.15018v1 | Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions | {
"content": "## Abstract\n\nAbstract Multiple-choice question answering (MCQA) is a key competence of performant transformer language models that is tested by mainstream benchmarks. However, recent evidence shows that models can have quite a range of performance, particularly when the task format is diversified slig... | [
{
"id": "NxNQMAEpco",
"initial_rating": 8,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "The paper is devoted to the analysis of the internal processing of Transformer LLMs when answering Multiple Choice Question Answering task. First, it is shown tha... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "3;4;3;4",
"confidence_avg": 3.5,
"soundness": "2;3;4;4",
"soundness_avg": 3.25,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;4;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.908201"
} | {
"id": "3OIMIQJ5np",
"metareview": "This paper investigates how LLMs solve Multiple Choice Question Answering (MCQA) tasks through mechanistic interpretability analysis. The key findings show that specific middle transformer layers and sparse attention heads play critical roles in selecting correct answers, with s... | {
"decision": "Accept (Spotlight)"
} |
6RtRsg8ZV1 | 2410.08896v1 | MAD-TD: Model-Augmented Data stabilizes High Update Ratio RL | {
"content": "## Abstract\n\nAbstract Building deep reinforcement learning (RL) agents that find a good policy with few samples has proven notoriously challenging.\nTo achieve sample efficiency, recent work has explored updating neural networks with large numbers of gradient steps for every new sample.\nWhile such hi... | [
{
"id": "nkanSXbqkx",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper addresses the challenge of unstable training in off-policy reinforcement learning (RL) methods when the update-to-data ratio is high. The authors ident... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "3;3;4;2",
"confidence_avg": 3,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;2;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.908944"
} | {
"id": "cIVnM5Mtq1",
"metareview": "This paper addresses the issue of overestimation in value estimates that preclude high update-to-data ratio. The paper proposes to use a world model and update the value using on-policy data generated by the world model to correct this overestimation. The effectiveness of their ... | {
"decision": "Accept (Spotlight)"
} |
6TLdqAZgzn | 2410.08208v2 | SPA: 3D Spatial-Awareness Enables Effective Embodied Representation | {
"content": "## Abstract\n\nAbstract In this paper, we introduce SPA, a novel representation learning framework that emphasizes the importance of 3D spatial awareness in embodied AI. Our approach leverages differentiable neural rendering on multi-view images to endow a vanilla Vision Transformer (ViT) with intrinsic... | [
{
"id": "mPkPhJcvKf",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper propose a representation learning framework named SPA to incopriate the 3D spatial awarness in embodied AI. SPA represents an advancement in embodied r... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "3;4;3;4",
"confidence_avg": 3.5,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"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": "2026-01-14T22:16:00.911040"
} | {
"id": "BkxYJIgRUk",
"metareview": "The paper introduces SPA, a framework that integrates 3D spatial awareness into Vision Transformers (ViTs) for embodied AI tasks. SPA uses differentiable neural rendering on multi-view images to improve the ViT’s understanding of 3D spatial relationships. The authors evaluate SP... | {
"decision": "Accept (Poster)"
} |
6UD3vymUst | 2408.05008v3 | FLOWDREAMER: EXPLORING HIGH FIDELITY TEXT-TO-3D GENERATION VIA RECTIFIED FLOW | {
"content": "## Abstract\n\nAbstract Recent advances in text-to-3D generation have made significant progress. In particular, with the pretrained diffusion models, existing methods predominantly use Score Distillation Sampling (SDS) to train 3D models such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3... | [
{
"id": "7Sg3ROYdkY",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This study proposes FlowDreamer to leverage a pretrained text-to-image (T2I) models trained via the rectified flow framework for score distillation sampling. Afte... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "4;4;5;4;4",
"confidence_avg": 4.2,
"soundness": "2;2;2;2;3",
"soundness_avg": 2.2,
"contribution": "2;1;2;2;3",
"contribution_avg": 2,
"presentation": "2;3;3;2;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.911758"
} | {
"id": "Jlu3514d0C",
"metareview": "This work introduces FlowDreamer, a novel approach that leverages pretrained text-to-image (T2I) models within the rectified flow framework for score distillation sampling (SDS). After detailing the training objective of SDS with a flow model, referred to as VFDS, the paper iden... | {
"decision": "Reject"
} |
6Vl9Uvxocp | 2402.02186v1 | Evolution guided generative flow networks | {
"content": "## Abstract\n\nAbstract Generative Flow Networks ( GFlowNets )are a family of probabilistic generative models that learn to sample compositional objects proportional to their rewards. One big challenge of GFlowNets is training them effectively when dealing with long time horizons and sparse rewards. To ... | [
{
"id": "VelyBzTmyO",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes using an Evolutionary Algorithm to fill a Prioritized Replay Buffer with (more) diverse trajectories to enhance the training process of Generat... | {
"rating": "3;3;5",
"rating_avg": 3.6666666666666665,
"confidence": "3;4;3",
"confidence_avg": 3.3333333333333335,
"soundness": "2;1;3",
"soundness_avg": 2,
"contribution": "2;2;2",
"contribution_avg": 2,
"presentation": "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": "2026-01-14T22:16:00.912452"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
6Vx28LSR7f | 2406.00622v1 | Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answering | {
"content": "## Abstract\n\nAbstract For vision-language models (VLMs), understanding the dynamic properties of objects and their interactions within 3D scenes from video is crucial for effective reasoning. Cognitive science research has suggested that humans are adept at understanding these properties by constructi... | [
{
"id": "C11ZWQLsKS",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper introduces DynSuperCLEVR, a video question answering (VideoQA) dataset that emphasizes understanding dynamic 3D object properties within 4D (3D + time) ... | {
"rating": "3;6;6;6",
"rating_avg": 5.25,
"confidence": "3;3;4;2",
"confidence_avg": 3,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.913135"
} | {
"id": "dkeIVLmlnr",
"metareview": "This work proposes a new VideoQA dataset that emphasizes understanding dynamic 3D object properties within 4D. All reviewers consistently recommended accepting this work. AC agrees that this work is interesting and deserves to be published on ICLR 2O25. The reviewers did raise s... | {
"decision": "Accept (Poster)"
} |
6X7HaOEpZS | 2410.16443v1 | Improving Neuron-level Interpretability with White-box Language Models | {
"content": "## Abstract\n\nAbstract Neurons in auto-regressive language models like GPT-2 can be interpreted by analyzing their activation patterns. Recent studies have shown that techniques such as dictionary learning, a form of post-hoc sparse coding, enhance this neuron-level interpretability.\nIn our research, ... | [
{
"id": "cG2v1HJpCn",
"initial_rating": 3,
"confidence": 3,
"soundness": 1,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes CRATE language models, a new alternative architecture to transformers that aims to be more inherently interpretable by encouraging a sparse/di... | {
"rating": "3;5;5",
"rating_avg": 4.333333333333333,
"confidence": "3;3;3",
"confidence_avg": 3,
"soundness": "1;3;3",
"soundness_avg": 2.3333333333333335,
"contribution": "2;2;2",
"contribution_avg": 2,
"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": "2026-01-14T22:16:00.913963"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
6XUSDvBFkV | 2408.01803v2 | STBLLM: Breaking the 1-Bit Barrier with Structured Binary LLMs | {
"content": "## Abstract\n\nAbstract In this paper, we present the first structural binarization method for LLM compression to less than 1-bit precision. Although LLMs have achieved remarkable performance, their memory-bound nature during the inference stage hinders the adoption of resource-constrained devices. Redu... | [
{
"id": "Gp0bSq0XZ6",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This work presents a structural binarization method for LLMs by combining N:M sparsity, residual approximation, and block-wise error compensation. Extensive exper... | {
"rating": "3;5;5;8",
"rating_avg": 5.25,
"confidence": "4;2;4;4",
"confidence_avg": 3.5,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.914684"
} | {
"id": "CAvHi7W3Eq",
"metareview": "The paper introduces STBLLM, a new method for compressing large language models (LLMs) to less than 1-bit precision, i.e. the paper's main claim. The main strength of the paper lies in its structured binarisation method, which employs N:M sparsity and fine-grained weight quantis... | {
"decision": "Accept (Poster)"
} |
6XodKiDS3B | 2410.22695v1 | Permutation Invariant Learning with High-Dimensional Particle Filters | {
"content": "## Abstract\n\nAbstract Sequential learning in deep models often suffers from challenges such as catastrophic forgetting and loss of plasticity, largely due to the permutation dependence of gradient-based algorithms, where the order of training data impacts the learning outcome. In this work, we introdu... | [
{
"id": "vAu6evOmgY",
"initial_rating": 3,
"confidence": 3,
"soundness": 1,
"contribution": 2,
"presentation": 3,
"summary": "In sequential learning, the permutation dependency of prevalent optimization algorithms such as gradient descent might suffer from catastrophic overfitting -- sev... | {
"rating": "3;3;6;8",
"rating_avg": 5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;1;3;4",
"soundness_avg": 2.5,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "1;3;2;4",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.915404"
} | {
"id": "3G1zJx9ros",
"metareview": "The paper tackles the problem of dependence of training on batch ordering in machine learning. In particular, the authors argue that issues such as forgetting and loss of plasticity can be addressed if the training is invariant to batch ordering. The authors note that the true B... | {
"decision": "Reject"
} |
6aHUmotXaw | 2408.06195v1 | Mutual Reasoning Makes Smaller LLMs Stronger Problem-Solver | {
"content": "## Abstract\n\nAbstract This paper introduces rStar, a self-play mutual reasoning approach that significantly improves reasoning capabilities of small language models (SLMs) without fine-tuning or superior models. rStar decouples reasoning into a self-play mutual generation-discrimination process. First... | [
{
"id": "5fOJIItqv2",
"initial_rating": 3,
"confidence": 2,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "The paper puts forward an approach to improve reasoning capabilities of LLMs. The approach is based on MCTS with a discriminator model selecting the most promisin... | {
"rating": "3;6;6;8",
"rating_avg": 5.75,
"confidence": "2;4;3;4",
"confidence_avg": 3.25,
"soundness": "1;3;2;4",
"soundness_avg": 2.5,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.916036"
} | {
"id": "3kzKrIfEcd",
"metareview": "(a) Summary:\nThe paper introduces rStar, a novel self-play mutual reasoning approach that improves small language models' (SLMs) reasoning capabilities without requiring fine-tuning or supervision from larger models. The key technical contributions include:\n1. A generation-dis... | {
"decision": "Accept (Poster)"
} |
6awxwQEI82 | 2410.03601v1 | How Discrete and Continuous Diffusion Meet: Comprehensive Analysis of Discrete Diffusion Models via a Stochastic Integral Framework | {
"content": "## Abstract\n\nAbstract Discrete diffusion models have gained increasing attention for their ability to model complex distributions with tractable sampling and inference. However, the error analysis for discrete diffusion models remains less well-understood.\nIn this work, we propose a comprehensive fra... | [
{
"id": "GmDDrCPKSk",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper addresses the challenge of error analysis in discrete diffusion\nmodels. To bridge the gap between discrete and continuous diffusion models\n(for which... | {
"rating": "6;6;8;8",
"rating_avg": 7,
"confidence": "2;2;2;3",
"confidence_avg": 2.25,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.917287"
} | {
"id": "Km1hHGBE5q",
"metareview": "In order to unify the theoretical analysis of continuous and discrete diffusion models, this paper introduces a stochastic integral formulation based on Lévy-type stochastic integrals and generalizes the Poisson random measure to one with time-independent and state-dependent int... | {
"decision": "Accept (Poster)"
} |
6bDJ3CIm5w | 2402.07322v1 | Interference Among First-Price Pacing Equilibria: A Bias and Variance Analysis | {
"content": "## Abstract\n\nAbstract. Online A/B testing is widely used in the internet industry to inform decisions on new feature roll-outs. For online marketplaces (such as advertising markets), standard approaches to A/B testing may lead to biased results when buyers operate under a budget constraint, as budget ... | [
{
"id": "zipMdO9qBG",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper considers A/B testing in online marketplaces, where interference arises because items can be recommended to advertisers in both the control and the tre... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "3;3;3",
"confidence_avg": 3,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "3;2;3",
"contribution_avg": 2.6666666666666665,
"presentation": "2;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.918974"
} | {
"id": "aUfNKcBJuP",
"metareview": "This paper studies A/B testing with interference in online marketplaces. The authors considers first-price pacing equilibrium (FPPE) and analyzes how the equilibrium changes according to contamination/interference. The authors then proposes a debias surrogate against the first-o... | {
"decision": "Accept (Poster)"
} |
6cGKi7FqJS | 2406.04321v2 | VidMuse: A Simple Video-to-Music Generation Framework with Long-Short-Term Modeling | {
"content": "## Abstract\n\nAbstract In this work, we systematically study music generation conditioned solely on the video.\nFirst, we present a large-scale dataset by collecting 360K video-music pairs, including various genres such as movie trailers, advertisements, and documentaries.\nFurthermore, we propose VidM... | [
{
"id": "v2sc1ZdLXb",
"initial_rating": 3,
"confidence": 5,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes Vidmuse, a video-to-music generation framework that generates high-fidelity music in sync with visual content. The authors also propose a larg... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;5;5;4",
"confidence_avg": 4.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "3;1;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.920076"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
6embY8aclt | 2410.13080v1 | Graph-constrained Reasoning: Faithful Reasoning on Knowledge Graphs with Large Language Models | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have demonstrated impressive reasoning abilities, but they still struggle with faithful reasoning due to knowledge gaps and hallucinations. To address these issues, knowledge graphs (KGs) have been utilized to enhance LLM reasoning through their struc... | [
{
"id": "k8fEcUXQW6",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper proposes graph-constrained reasoning (GCR). GCR integrates KG structure into the LLM decoding process through KG-Trie, a trie-based index that encodes ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"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": "2026-01-14T22:16:00.920859"
} | {
"id": "jpwfMSpoyH",
"metareview": "Graph-constrained reasoning (GCR) is introduced to integrate knowledge graph (KG) structure into the LLM decoding process. It leverages a KG-specialized LLM for graph-constrained reasoning and a general LLM for inductive reasoning over multiple reasoning paths. To alleviate hall... | {
"decision": "Reject"
} |
6guG2OlXsr | 2410.11710v1 | MTU-Bench: A Multi-granularity Tool-Use Benchmark for Large Language Models | {
"content": "## Abstract\n\nAbstract Large Language Models (LLMs) have displayed massive improvements in reasoning and decision-making skills and can hold natural conversations with users.\nRecently, many tool-use benchmark datasets have been proposed.\nHowever, existing datasets have the following limitations:\n(1)... | [
{
"id": "rCaM39f4X4",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes a multi-granularity tool-use benchmark for large language models, called MTU-Bench.\n\nThe main contribution of this paper can be summarized a... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "3;2;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.921591"
} | {
"id": "RIOCZrp1Wu",
"metareview": "The paper introduced a new benchmark MTU-Bench for evaluating the LLM's abilities to use tools in multiple scenarios. MTU-Bench considers more granular settings in aspects of the number of tools that can use and the number of rounds of tools can use in multi-turn conversation. T... | {
"decision": "Accept (Poster)"
} |
6iM7mmVhXh | 2408.05098v1 | Exploring the Limitations of Layer Synchronization in Spiking Neural Networks | {
"content": "## Abstract\n\nAbstract Currently, neural-network processing in machine learning applications relies on layer synchronization, whereby neurons in a layer aggregate incoming currents from all neurons in the preceding layer, before evaluating their activation function. This is practiced even in artificial... | [
{
"id": "nbHpbuqlRZ",
"initial_rating": 6,
"confidence": 3,
"soundness": 4,
"contribution": 3,
"presentation": 2,
"summary": "Current spike neural networks (SNNs) must compute and integrate all presynaptic currents from the previous layer before performing calculations for the neurons in... | {
"rating": "3;5;6;8",
"rating_avg": 5.5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.922289"
} | {
"id": "4uCSJ11D3z",
"metareview": "This paper examines the possibility of using asynchronous processing for spiking neural networks (SNNs). In brief, the authors are interested in eliminating the constraint that is often imposed in multi-layer SNNs of having every cell on a layer accumulate its inputs and spike (... | {
"decision": "Reject"
} |
6jyEj4rGZJ | 2409.08520v2 | GroundingBooth: Grounding Text-to-Image Customization | {
"content": "## Abstract\n\nAbstract Recent studies in text-to-image customization show great success in generating personalized object variants given several images of a subject. While existing methods focus more on preserving the identity of the subject, they often fall short of controlling the spatial relationshi... | [
{
"id": "VQAAo5vp86",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper proposes a framework which allows users to customize an image by 1) specifying the position (layout) of the object, and 2) providing a reference image ... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "4;3;3;3;4",
"confidence_avg": 3.4,
"soundness": "3;2;3;2;3",
"soundness_avg": 2.6,
"contribution": "2;2;3;2;3",
"contribution_avg": 2.4,
"presentation": "4;3;3;2;2",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.922927"
} | {
"id": "FVC2Kukrbr",
"metareview": "This paper proposes a grounded text-to-image generation framework incorporating reference objects and bounding-box constraints. While backed by comprehensive experiments, in private discussion period, the reviewers find the contributions incremental. The core techniques, such a... | {
"decision": "Reject"
} |
6kjTRMJ3be | 2410.02026v1 | Zodiac: A Cardiologist-Level LLM Framework for Multi-Agent Diagnostics | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) have demonstrated remarkable progress in healthcare. However, a significant gap remains regarding LLMs’ professionalism in domain-specific clinical practices, limiting their application in real-world diagnostics. In this work, we introduce Zodiac , an... | [
{
"id": "RfUaqR9oW7",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents ZODIAC, a human-in-the-loop multi-agent framework for electrocardiograms (ECGs) that leverages real-world patient data and cardiologist-adjudi... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "1;2;3;2",
"soundness_avg": 2,
"contribution": "1;2;3;2",
"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": "2026-01-14T22:16:00.923532"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
6ktqrC1Bpf | 2410.19110v1 | bio2token: all-atom tokenization of any biomolecular structure with mamba | {
"content": "## \\thesubsection Tokenizer results\n\n### \\thesubsubsection All-domain tokenizing: bio2token\n\n**Table \\thetable: bio2token results: The reconstruction error is the atom-wise rmse between the ground truth structure point cloud and the reconstructed point cloud from the tokens. ”bb” and ”sc” are the... | [
{
"id": "PNuqTEi0cf",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes to train a mamba-based auto-encoder on biomolecular structures to allow for accurate tokenization (i.e. conversion to discrete tokens). The au... | {
"rating": "3;3;5;5;6",
"rating_avg": 4.4,
"confidence": "5;4;4;3;3",
"confidence_avg": 3.8,
"soundness": "2;3;4;2;3",
"soundness_avg": 2.8,
"contribution": "3;2;2;2;3",
"contribution_avg": 2.4,
"presentation": "2;2;3;3;3",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.924090"
} | {
"id": "PSjawpOzv7",
"metareview": "(a) The paper proposes a method for all-atom tokenization of biomolecular structures using a quantized auto-encoder with the Mamba state space model. It claims to achieve high reconstruction accuracies for proteins, RNA, and small molecules. The method is shown to be scalable to... | {
"decision": "Reject"
} |
6ldD8Y4gBQ | 2410.09101v1 | Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning | {
"content": "## Abstract\n\nAbstract Dataset ownership verification, the process of determining if a dataset is used in a model’s training data, is necessary for detecting unauthorized data usage and data contamination.\nExisting approaches, such as backdoor watermarking, rely on inducing a detectable behavior into ... | [
{
"id": "MgDn1nEfVt",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces data taggants, a novel non-backdoor dataset ownership verification technique that helps detect if machine learning models were trained using... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;5;4",
"confidence_avg": 4.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.924687"
} | {
"id": "P0cKhwd6hO",
"metareview": "The submission \"Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning\" proposes a dataset attribution via watermarking method using clean-label data poisoning. While reviewers point out that the exact algorithm used for clean-label poisoning is no... | {
"decision": "Accept (Poster)"
} |
6nZwOYDcQx | 2408.10280v2 | NoRA: Nested Low-Rank Adaptation for Efficient Fine-Tuning Large Models | {
"content": "## Abstract\n\nAbstract In this paper, we introduce Nested Low-Rank Adaptation (NoRA), a novel approach to parameter-efficient fine-tuning that extends the capabilities of Low-Rank Adaptation (LoRA) techniques. Vanilla LoRA overlooks pre-trained weight inheritance and still requires fine-tuning numerous... | [
{
"id": "a8BQElLffx",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper titled \"NORA: NESTED LOW-RANK ADAPTATION FOR EFFICIENT FINE-TUNING LARGE MODELS\" introduces a novel parameter-efficient fine-tuning method named NoRA ... | {
"rating": "3;3;3;5;6",
"rating_avg": 4,
"confidence": "4;5;4;4;2",
"confidence_avg": 3.8,
"soundness": "2;1;2;3;4",
"soundness_avg": 2.4,
"contribution": "2;1;2;3;3",
"contribution_avg": 2.2,
"presentation": "2;1;2;3;4",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.925303"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
6oWFn6fY4A | 2403.14715v2 | Towards Understanding Why Label Smoothing Degrades Selective Classification and How to Fix It | {
"content": "## Abstract\n\nAbstract Label smoothing (LS) is a popular regularisation method for training neural networks as it is effective in improving test accuracy and is simple to implement. “Hard” one-hot labels are “smoothed” by uniformly distributing probability mass to other classes, reducing overfitting.\n... | [
{
"id": "cHuYSziZs2",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper investigates the impact of label smoothing (LS) on selective classification (SC), showing that while LS is a popular regularization technique for impro... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;2",
"soundness_avg": 2.75,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.926063"
} | {
"id": "bhSGcjYkk7",
"metareview": "The paper studies the effect of label smoothing (LS) on selective classification (SC) performance. Empirically, it is shown that LS systematically degrades the performance of the maximum softmax probability SC baseline. Analytically, it is shown that the reason for this can be t... | {
"decision": "Accept (Poster)"
} |
6ozaf7VRIP | 2407.04973v1 | LogicVista: Multimodal LLM Logical Reasoning Benchmark in Visual Contexts | {
"content": "## Abstract\n\nAbstract We propose LogicVista, an evaluation benchmark that assesses the integrated logical reasoning capabilities of multimodal large language models (MLLMs) in Vis ual contexts. Recent advancements in MLLMs have demonstrated various fascinating abilities, from crafting poetry based on ... | [
{
"id": "MQ9KJlGnju",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper presents LogicVista, a benchmark for evaluating MLLMs' logical reasoning abilities. It includes 448 multi-choice questions spanning 11 abilities, with ... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "4;4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;3;3;2",
"soundness_avg": 2.4,
"contribution": "2;3;3;2;2",
"contribution_avg": 2.4,
"presentation": "3;2;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.926914"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
6qUUgw9bAZ | 2410.04707v1 | Learning How Hard to Think: Input-Adaptive Allocation of LM Computation | {
"content": "## Abstract\n\nAbstract Computationally intensive decoding procedures—including search, reranking, and self-critique—can improve the quality of language model (LM) outputs in problems spanning code generation, numerical reasoning, and dialog.\nExisting work typically applies the same decoding procedure ... | [
{
"id": "n0lyTxeTcz",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents an approach to adaptively allocate computational resources for language model (LM) decoding based on input difficulty. The authors propose a f... | {
"rating": "5;5;5;8",
"rating_avg": 5.75,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;2;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.927941"
} | {
"id": "wqcLJgfeyf",
"metareview": "The paper presents an adaptive computation allocation approach for LM decoding that predicts input difficulty to optimize resource usage. Key strengths include: comprehensive evaluation across diverse domains (code, math, chat), significant compute reduction (up to 50%) without ... | {
"decision": "Accept (Poster)"
} |
6rMHcLWxl4 | 2410.05363v1 | Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation | {
"content": "## Abstract\n\nAbstract Text-to-video (T2V) models like Sora have made significant strides in visualizing complex prompts, which is increasingly viewed as a promising path towards constructing the universal world simulator. Cognitive psychologists believe that the foundation for achieving this goal is t... | [
{
"id": "d0XBO5Em8S",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "Although T2V models have shown great progress in generating good media-level content, this paper challenges their capability to become the real world simulator. T... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "4;4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;2;2;3",
"soundness_avg": 2.2,
"contribution": "2;2;2;4;3",
"contribution_avg": 2.6,
"presentation": "2;3;4;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.928654"
} | {
"id": "xAs0LIe8HV",
"metareview": "The submission introduces a new benchmark for assessing models' understanding of physical commonsense. Reviewers were lukewarm about the submission, and all shared concerns about the heavy use of generative AI during benchmark development. Other concerns include insufficient ev... | {
"decision": "Reject"
} |
6tyPSkshtF | 2410.07574v1 | Gap-Dependent Bounds for Q-Learning using Reference-Advantage Decomposition | {
"content": "## Abstract\n\nAbstract We study the gap-dependent bounds of two important algorithms for on-policy Q 𝑄 Q italic_Q -learning for finite-horizon episodic tabular Markov Decision Processes (MDPs): UCB-Advantage (Zhang et al. 2020) and Q-EarlySettled-Advantage (Li et al. 2021). UCB-Advantage and Q-EarlySe... | [
{
"id": "9j7csRvwjj",
"initial_rating": 8,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper establishes improved gap-dependent upper bounds on finite-horizon episodic Markov decision processes (MDPs). There already exists a gap-dependent upper... | {
"rating": "5;6;8;8",
"rating_avg": 6.75,
"confidence": "4;3;3;2",
"confidence_avg": 3,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3;3",
"contribution_avg": 3,
"presentation": "2;2;3;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.930652"
} | {
"id": "Mkq5dVCd2H",
"metareview": "This submission studies gap-dependent bounds and policy switching cost for Q learning algorithms which use variance estimation. \n\nThis paper gives novel gap-dependent variance-aware regret bounds, and provides an algorithm with a gap-dependent policy switching cost. These theo... | {
"decision": "Accept (Spotlight)"
} |
6z4YKr0GK6 | 2410.05080v2 | ScienceAgentBench: Toward Rigorous Assessment of Language Agents for Data-Driven Scientific Discovery | {
"content": "## Abstract\n\nAbstract The advancements of language language models (LLMs) have piqued growing interest in developing LLM-based language agents to automate scientific discovery end-to-end, which has sparked both excitement and skepticism about their true capabilities. In this work, we call for rigorous... | [
{
"id": "6MpSK3S21A",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces ScienceAgentBench, a new benchmark designed to test how well language agents can handle tasks in data-driven scientific discovery. The autho... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "5;3;4",
"confidence_avg": 4,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "3;3;3",
"contribution_avg": 3,
"presentation": "2;3;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.931896"
} | {
"id": "VgXBiw6AQQ",
"metareview": "We recommend the paper to be accepted for Poster.\n\nThe contribution seems timely and addresses some concerns in the literature. \n\nBelow more details about this contribution. \n\nThe paper introduces ScienceAgentBench, a new benchmark for evaluating language agents for data-d... | {
"decision": "Accept (Poster)"
} |
71XtUhazG0 | 2408.02034v3 | Mini-Monkey: Alleviating the Semantic Sawtooth Effect for Lightweight MLLMs via Complementary Image Pyramid | {
"content": "## Abstract\n\nAbstract Recently, scaling images to high resolution has received much attention in multimodal large language models (MLLMs). Most existing practices adopt a sliding-window-style cropping strategy to adapt to resolution increase. Such a cropping strategy, however, can easily cut off objec... | [
{
"id": "HJM3K8IsaL",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper introduces the \"semantic sawtooth effect\" caused by common cropping strategies in high-resolution image scaling for MLLMs. To tackle this issue, they ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;3;3;2",
"contribution_avg": 2.5,
"presentation": "2;3;2;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.932646"
} | {
"id": "G49Ws9ku61",
"metareview": "This paper tackles the high resolution image requirement in MLLM, and propose a simple, plug-and play method named Complementary Image Pyramid (CIP) to mitigate the semantic sawtooth effect of patch tileing. The reviewers approve for the effectiveness of the proposed method, and... | {
"decision": "Accept (Poster)"
} |
72OSO38a2z | 2410.01295v1 | LaGeM: A Large Geometry Model for 3D Representation Learning and Diffusion | {
"content": "## Abstract\n\nAbstract This paper introduces a novel hierarchical autoencoder that maps 3D models into a highly compressed latent space. The hierarchical autoencoder is specifically designed to tackle the challenges arising from large-scale datasets and generative modeling using diffusion. Different fr... | [
{
"id": "cZth2ENmnS",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a novel hierarchical autoencoder that compresses 3D models into a highly compact latent space, designed to handle large-scale datasets and s... | {
"rating": "5;6;6;8",
"rating_avg": 6.25,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "2;2;3;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": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.933239"
} | {
"id": "IXYpCgnCv1",
"metareview": "The paper introduces a hierarchical autoencoder for 3D shapes that allows to train diffusion models in for generation in latent space.\nThe paper was well-received by all reviewers, converging to positive scores, recommending acceptance. The reviewers highlighted the good writin... | {
"decision": "Accept (Poster)"
} |
72nCh5JtLQ | 2410.10112v1 | Can We Predict Performance of Large Models across Vision-Language Tasks? | {
"content": "## Abstract\n\nAbstract Evaluating large vision-language models (LVLMs) is very expensive, due to the high computational costs and the wide variety of tasks. The good news is that if we already have some observed performance scores, we may be able to infer unknown ones. In this study, we propose a new f... | [
{
"id": "zibk6XWlm3",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper provides a framework for predicting the performance of large vision-language models on held-out downstream tasks using a small set of observed task per... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "3;3;5;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "3;2;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.933919"
} | {
"id": "N81wCwzN3i",
"metareview": "This paper proposes a new framework for predicting unknown performance scores based on observed ones from other LVLMs or tasks. This paper formulates the performance prediction as a matrix completion task and applies probabilistic matrix factorization with Markov chain Monte Car... | {
"decision": "Reject"
} |
72yPbvSx0c | 2312.01544v2 | Koopman Embedded Equivariant Control | {
"content": "## Abstract\n\nAbstract This paper investigates how representation learning can enable optimal control in unknown and complex dynamics, such as chaotic and non-linear systems, without relying on prior domain knowledge of the dynamics. The core idea is to establish an equivariant geometry that is diffeom... | [
{
"id": "lgl3iH4jsC",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a method for solving control problems called Koopman embedded equivariant control (KEEC). The key idea of the paper is that the state of the d... | {
"rating": "3;3;6;6",
"rating_avg": 4.5,
"confidence": "4;4;2;2",
"confidence_avg": 3,
"soundness": "2;3;3;2",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"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": "2026-01-14T22:16:00.934748"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
74QmBTV0Zf | 2409.04701v2 | Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding Models | {
"content": "## Abstract\n\nAbstract Many use cases require retrieving smaller portions of text, and dense vector-based retrieval systems often perform better with shorter text segments, as the semantics are less likely to be “over-compressed” in the embeddings.\nConsequently, practitioners often split text document... | [
{
"id": "huL3qylUfW",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper introduces late chunking for document embeddings, which suggests that instead of chunking the text and then computing the embedding for individual chunk... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "1;3;2;3",
"contribution_avg": 2.25,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.936807"
} | {
"id": "OGGztFT6ox",
"metareview": "The paper proposes a latent chunking approach for contextualized chunk representation. \n\nReviewers generally gave borderline or rejection scores. Several major concerns are related to experimental setups and lack of evaluation on downstream tasks. Even the reviewer who gave a ... | {
"decision": "Reject"
} |
74vnDs1R97 | 2406.07506v1 | Understanding Visual Concepts Across Models | {
"content": "## Abstract\n\nAbstract Large multimodal models such as Stable Diffusion can generate, detect, and classify new visual concepts after fine-tuning just a single word embedding. Do models learn similar words for the same concepts (i.e. <orange-cat> = orange + cat)? We conduct a large-scale analysis on thr... | [
{
"id": "Ny5Ja4Lrxw",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper examines how fine-tuned prompt embeddings for visual concepts affect text-to-image generation, object detection, and classification. It reveals that th... | {
"rating": "5;5;5;5;6",
"rating_avg": 5.2,
"confidence": "4;3;1;3;3",
"confidence_avg": 2.8,
"soundness": "3;2;2;2;3",
"soundness_avg": 2.4,
"contribution": "2;3;1;2;3",
"contribution_avg": 2.2,
"presentation": "3;3;3;2;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.937483"
} | {
"id": "Iuc5tkDtQU",
"metareview": "The paper examines prompt embeddings in large multimodal models like Stable Diffusion, revealing that optimized prompts (via prompt-tuning) resemble adversarial attacks on text encoders. Analyzing 4,800 embeddings across tasks, the study shows these perturbations target the fina... | {
"decision": "Accept (Poster)"
} |
7893vsQenk | 2410.04682v2 | On the Adversarial Risk of Test Time Adaptation: An Investigation into Realistic Test-Time Data Poisoning | {
"content": "## Abstract\n\nAbstract Test-time adaptation (TTA) updates the model weights during the inference stage using testing data to enhance generalization. However, this practice exposes TTA to adversarial risks. Existing studies have shown that when TTA is updated with crafted adversarial test samples, also ... | [
{
"id": "4mrR52eqsW",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper investigates adversarial risks in test-time adaptation (TTA), which updates model weights during inference to counter distribution shifts. The authors ... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;3;3",
"confidence_avg": 3.5,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "2;2;2;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.938225"
} | {
"id": "Ud7eKNKWIo",
"metareview": "\"On the Adversarial Risk of Test Time Adaptation: An Investigation into Realistic Test-Time Data Poisoning\" re-investigates assumptions made in previous work on data poisoning during test time augmentation, propose a new gray-box setting and provide a careful investgation in t... | {
"decision": "Accept (Poster)"
} |
78NPsEq8cF | 2406.02539v2 | Parrot: Multilingual Visual Instruction Tuning | {
"content": "## Abstract\n\nAbstract The rapid development of Multimodal Large Language Models (MLLMs) like GPT-4V has marked a significant step towards artificial general intelligence.\nExisting methods mainly focus on aligning vision encoders with LLMs through supervised fine-tuning (SFT) to endow LLMs with multim... | [
{
"id": "WdtI2l2ik8",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 4,
"presentation": 4,
"summary": "This paper aims at the multilingual issues in recent MLLMs and proposes a MoE-based alignment layer to this end. First, the authors find that existing MLLMs are n... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;3;4",
"contribution_avg": 2.75,
"presentation": "3;3;3;4",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.942164"
} | {
"id": "IS4GZbuyhV",
"metareview": "(a) Summary:\nThe paper introduces Parrot, a Mixture-of-Experts (MoE)-based model to improve multilingual alignment in Multimodal Large Language Models (MLLMs). It addresses imbalances in non-English data and proposes MMMB, a multilingual benchmark. Empirical results demonstrate... | {
"decision": "Reject"
} |
78Nn4QJTEN | 2410.10781v1 | When Attention Sink Emerges in Language Models: An Empirical View | {
"content": "## Abstract\n\nAbstract Language Models (LMs) assign significant attention to the first token, even if it is not semantically important, which is known as attention sink . This phenomenon has been widely adopted in applications such as streaming/long context generation, KV cache optimization, inference ... | [
{
"id": "uFFuRvlKnP",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 3,
"summary": "The paper investigates the phenomenon of attention sink in LMs, where significant attention is allocated to the first token regardless of its semantic importance.... | {
"rating": "6;8;8",
"rating_avg": 7.333333333333333,
"confidence": "4;5;4",
"confidence_avg": 4.333333333333333,
"soundness": "4;4;4",
"soundness_avg": 4,
"contribution": "3;4;3",
"contribution_avg": 3.3333333333333335,
"presentation": "3;3;4",
"presentation_avg": 3.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.942924"
} | {
"id": "xRx946Ild0",
"metareview": "This paper explores the phenomenon of attention sink in language models (LMs). Attention sink describes how, in autoregressive Transformer-based LMs, a disproportionate amount of attention often gets allocated to the first token in the sequence, regardless of its semantic import... | {
"decision": "Accept (Spotlight)"
} |
79fjGDmw90 | 2406.05343v2 | M3GIA: A Cognition Inspired Multilingual and Multimodal General Intelligence Ability Benchmark | {
"content": "## Abstract\n\nAbstract As recent multi-modality large language models (MLLMs) have shown formidable proficiency on various complex tasks, there has been increasing attention on debating whether these models could eventually mirror human intelligence.\nHowever, existing benchmarks mainly focus on evalua... | [
{
"id": "EjAH2nPuFz",
"initial_rating": 3,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper presents a benchmark M3GIA which claims to act as the first “IQ test” for multimodal large language models (MLLM). It is built based on five cognitive ... | {
"rating": "3;5;5",
"rating_avg": 4.333333333333333,
"confidence": "3;4;5",
"confidence_avg": 4,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "2;2;2",
"contribution_avg": 2,
"presentation": "2;2;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.943865"
} | {
"id": "8BJXQ0Hgk3",
"metareview": "The paper proposes a broad benchmark to test models in different cognitively inspired tasks across modalities and skills.\n\nStrengths: \nProvides reasoning for validity. \nRaises discussions\n\nWeaknesses:\nSmall number of examples (perhaps a reliability test may account for ... | {
"decision": "Reject"
} |
7ANDviElAo | 2405.14260v2 | Graph Sparsification via Mixture of Graphs | {
"content": "## Abstract\n\nAbstract Graph Neural Networks (GNNs) have demonstrated superior performance across various graph learning tasks but face significant computational challenges when applied to large-scale graphs. One effective approach to mitigate these challenges is graph sparsification, which involves re... | [
{
"id": "l56c9a1A0E",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper presents a novel graph sparsification method, Mixture-of-Graphs (MoG), to optimize Graph Neural Networks (GNNs) for large-scale graphs. MoG dynamically... | {
"rating": "5;6;6;6",
"rating_avg": 5.75,
"confidence": "4;2;3;4",
"confidence_avg": 3.25,
"soundness": "3;4;3;2",
"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 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.944804"
} | {
"id": "1j0UKkVDGQ",
"metareview": "In this submission, the authors leverage the idea of the mixture of graphs (MoG) to achieve adaptive graph sparsification. Applying MoG to GNNs can reduce the computational costs significantly while maintaining high performance on large-scale graph learning tasks. The idea of Mo... | {
"decision": "Accept (Spotlight)"
} |
7DY2Nk9snh | 2402.01832v2 | SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training? | {
"content": "## Abstract\n\nAbstract We present SynthCLIP, a CLIP model trained on entirely synthetic text-image pairs. Leveraging recent text-to-image (TTI) networks and large language models (LLM), we generate synthetic datasets of\nimages and corresponding captions at scale, with no human intervention. In this wo... | [
{
"id": "Qv1CfXZWT8",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper explores the performance of CLIP-style models trained on purely synthetic image-caption pairs (called SynthCLIP) generated by modern text-to-image diffu... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"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": "2026-01-14T22:16:00.945647"
} | {
"id": "LYTiWQGHC3",
"metareview": "This paper discusses using pure synthetic data for training CLIP models, and shows the performance, scaling property, and analysis compared with real datasets. The paper also introduces a new SynthCI-30M with captions on 30 million images. However, the paper still lacks enough e... | {
"decision": "Reject"
} |
7Dub7UXTXN | 2406.12615v1 | When Are Bias-Free ReLU Networks Effectively Linear Networks? | {
"content": "## Abstract\n\nAbstract We investigate the expressivity and learning dynamics of bias-free ReLU networks. We firstly show that two-layer bias-free ReLU networks have limited expressivity: the only odd function two-layer bias-free ReLU networks can express is a linear one. We then show that, under symmet... | [
{
"id": "xKQfOOM5q3",
"initial_rating": 6,
"confidence": 4,
"soundness": 4,
"contribution": 3,
"presentation": 4,
"summary": "The paper studies what functions bias-free ReLU and leaky ReLU networks can represent and what the training dynamics are. A difference between two-layer networks ... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;4;3;4",
"confidence_avg": 3.75,
"soundness": "2;3;3;4",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "4;3;3;4",
"presentation_avg": 3.5
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.946435"
} | {
"id": "DY48LRFq1K",
"metareview": "The authors provide a theoretical analysis of expressivity and training dynamics of ReLU networks with no bias, and in particular compare it against a linear network. The most novel result being Theorem 7, where under some conditions, the gradient flow trajectories of a two laye... | {
"decision": "Reject"
} |
7E7v5mJnfl | 2406.00259v1 | PuzzleFusion++: Auto-agglomerative 3D Fracture Assembly by Denoise and Verify | {
"content": "## Abstract\n\nAbstract This paper proposes a novel “auto-agglomerative” 3D fracture assembly method, PuzzleFusion++, resembling how humans solve challenging spatial puzzles. Starting from individual fragments, the approach 1) aligns and merges fragments into larger groups akin to agglomerative clusteri... | [
{
"id": "KPusP2Oj3X",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes PuzzleFusion++, an \"auto-agglomerative\" method for the task of 3D fracture assembly.\nSpecifically, the proposed pipeline undergoes a iterat... | {
"rating": "5;6;6;6;8",
"rating_avg": 6.2,
"confidence": "5;5;4;4;5",
"confidence_avg": 4.6,
"soundness": "2;2;3;3;3",
"soundness_avg": 2.6,
"contribution": "2;2;3;3;3",
"contribution_avg": 2.6,
"presentation": "3;3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.947185"
} | {
"id": "5nLePCIUrc",
"metareview": "This paper proposes a 3D fracture assembly method that learns a diffusion model to predict a 6-DoF alignment for each fragment iteratively, followed by a transformer model that verifies and merges pairwise alignments into larger ones. Unlike previous diffusion-based models, it s... | {
"decision": "Accept (Poster)"
} |
7EK2hqWmvz | 2405.15198v2 | RAEE: A Robust Retrieval-Augmented Early Exit Framework for Efficient Inference | {
"content": "## Abstract\n\nAbstract Deploying large language model inference remains challenging due to their high computational overhead.\nEarly exiting optimizes model inference by adaptively reducing the number of inference layers.\nExisting methods typically train internal classifiers to determine whether to ex... | [
{
"id": "i1VGReUdGD",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This work uses retrieval to improve performance on training-free early exit frameworks. The motivation for doing so is observing that similar data should have sim... | {
"rating": "3;3;5;6",
"rating_avg": 4.25,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;1;2;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.947832"
} | {
"id": "Y1M5VwoX22",
"metareview": "This paper received mixed reviews. The reviewers recognized the new, interesting, and reasonable method, its strong performance, and extensive experiments. At the same time, they raised concerns with unclear motivation and inappropriate positioning of the paper (qFBa), no compar... | {
"decision": "Reject"
} |
7EhS3YBxjY | 2407.01509v3 | MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs | {
"content": "## Abstract\n\nAbstract We introduce MIA-Bench, a new benchmark designed to evaluate multimodal large language models (MLLMs) on their ability to strictly adhere to complex instructions. Our benchmark comprises a diverse set of 400 image-prompt pairs, each crafted to challenge the models’ compliance wit... | [
{
"id": "my2gSnQNCf",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces MIA-Bench, a benchmark designed to evaluate multimodal large language models (MLLMs) on strict adherence to complex instructions. Based on a... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.948470"
} | {
"id": "TyQchDw1g4",
"metareview": "This paper introduces a new MIA-Bench benchmark specifically designed for MLLM's complex instruction following ability study. The authors' major contribution is a new dataset containing 400 image-instruction pairs, where the instructions are written by humans with high quality. ... | {
"decision": "Accept (Poster)"
} |
7FQDHv9fD4 | 2407.19160v1 | Decomposing heterogeneous dynamical systems with graph neural networks | {
"content": "## Abstract\n\nAbstract Natural physical, chemical, and biological dynamical systems are often complex, with heterogeneous components interacting in diverse ways.\nWe show that graph neural networks can be designed to jointly learn the interaction rules and the structure of the heterogeneity from data a... | [
{
"id": "TCGxhEd22P",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes using graph neural networks (GNNs) to jointly learn interaction rules and heterogeneous structure in complex dynamical systems from data alone.... | {
"rating": "3;3;5;5",
"rating_avg": 4,
"confidence": "2;5;3;4",
"confidence_avg": 3.5,
"soundness": "2;1;3;3",
"soundness_avg": 2.25,
"contribution": "3;1;2;2",
"contribution_avg": 2,
"presentation": "2;1;3;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.949292"
} | {
"id": "RmegvylxaF",
"metareview": "The paper proposes a graph neural network framework for learning the dynamics of a physical system, which is tested on several simulations.\n\nWe had four reviews, all negative. All reviewers are unanimous concerning the lack of novelty, poor writing and presentation (e.g., Figu... | {
"decision": "Reject"
} |
7GKbQ1WT1C | 2403.08743v1 | Prompting Fairness: Integrating Causality to Debias Large Language Models | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) can easily generate biased and discriminative responses.\nAs LLMs tap into consequential decision-making (e.g., hiring and healthcare), it is of crucial importance to develop strategies to mitigate these biases.\nThis paper focuses on social bias, tac... | [
{
"id": "jlyvl0Wyr3",
"initial_rating": 5,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a prompt-based method to remove biases in a language model's output. It motivates the prompts using the idea of selection bias from causal i... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "3;5;4;2",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;3;2",
"contribution_avg": 2,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.950051"
} | {
"id": "lQ7wAyzAgS",
"metareview": "The paper presents a novel framework that leverages causal analysis to mitigate biases in large language models (LLMs) through prompting strategies. The paper’s strengths include its theoretical foundation, extensive experimental results demonstrating bias reduction, and a compr... | {
"decision": "Accept (Poster)"
} |
7H1jbTaOIn | 2408.00144v1 | Distributed In-Context Learning under Non-IID Among Clients | {
"content": "## Abstract\n\nAbstract Advancements in large language models (LLMs) have shown their effectiveness in multiple complicated natural language reasoning tasks.\nA key challenge remains in adapting these models efficiently to new or unfamiliar tasks.\nIn-context learning (ICL) provides a promising solution... | [
{
"id": "btgT84NOb9",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "This paper tackles the challenge of distributed non-IID in-context learning (ICL) for LLMs, where data is spread across clients with differing distributions. The ... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "4;4;4;4",
"confidence_avg": 4,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;3;2",
"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": "2026-01-14T22:16:00.950754"
} | {
"id": "CzkfWFORIe",
"metareview": "In this paper, the authors proposed a new setting for ICL, where training data is stored in a distributed manner. \n\nThere are some major concerns raised by the reviewer. 1, The assumption of the proposed method may not be practical. Though the authors added some experiments, t... | {
"decision": "Reject"
} |
7HEMpBTb3R | 2406.11608v1 | Visually Consistent Hierarchical Image Classification | {
"content": "## Abstract\n\nAbstract Hierarchical semantic classification requires the prediction of a taxonomy tree instead of a single flat level of the tree, where both accuracies at individual levels and consistency across levels matter. We can train classifiers for individual levels, which has accuracy but not ... | [
{
"id": "6rORmTsuOr",
"initial_rating": 6,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper proposes H-CAST architecture for hierarchical classification tasks. The architecture builds on top of prior CAST work: superpixels are fed into a ViT ne... | {
"rating": "3;3;5;6;8",
"rating_avg": 5,
"confidence": "4;5;5;4;4",
"confidence_avg": 4.4,
"soundness": "2;2;2;2;4",
"soundness_avg": 2.4,
"contribution": "2;1;2;2;3",
"contribution_avg": 2,
"presentation": "3;2;2;2;4",
"presentation_avg": 2.6
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.951371"
} | {
"id": "EOPeYY6dFC",
"metareview": "The paper proposes an architecture that extends CAST for hierarchical classification tasks. The proposal fuses superpixels of high similarity using a graph-pooling operation within the ViT tokens. The hierarchical classification is achieved through a set of classification heads ... | {
"decision": "Accept (Poster)"
} |
7IzeL0kflu | 2407.04811v2 | Simplifying Deep Temporal Difference Learning | {
"content": "## Abstract\n\nAbstract Q 𝑄 Q italic_Q -learning played a foundational role in the field reinforcement learning (RL).\nHowever, TD algorithms with off-policy data, such as Q 𝑄 Q italic_Q -learning, or nonlinear function approximation like deep neural networks require several additional tricks to stabi... | [
{
"id": "S2RsQSuwOU",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "Modern deep-reinforcement learning resorts to techniques such as replay buffers and target networks to provide stability with nonlinear off-policy learning. Howev... | {
"rating": "3;5;8;8",
"rating_avg": 6,
"confidence": "3;3;2;3",
"confidence_avg": 2.75,
"soundness": "3;3;3;4",
"soundness_avg": 3.25,
"contribution": "3;2;4;3",
"contribution_avg": 3,
"presentation": "1;2;4;3",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.952332"
} | {
"id": "EmYFH6Bss6",
"metareview": "This paper makes two contributions. First, a proof that TD learning converges when the network uses layer normalization and weight-decay. This is demonstrated in experiments that show that one need not use a target value network or a “replay buffer” (there was a lot of back and ... | {
"decision": "Accept (Spotlight)"
} |
7JlL8ECPJ7 | 2410.08336v1 | Kernel Banzhaf: A Fast and Robust Estimator for Banzhaf Values | {
"content": "## Abstract\n\nAbstract Banzhaf values offer a simple and interpretable alternative to the widely-used Shapley values. We introduce Kernel Banzhaf, a novel algorithm inspired by KernelSHAP, that leverages an elegant connection between Banzhaf values and linear regression. Through extensive experiments o... | [
{
"id": "ORLWmUAYp5",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "Inspired by KernelSHAP, the paper introduces a method named \"Kernel Banzhaf\" that connects Banzhaf values to linear regression, leveraging \"leverage score samp... | {
"rating": "5;6;8",
"rating_avg": 6.333333333333333,
"confidence": "3;3;4",
"confidence_avg": 3.3333333333333335,
"soundness": "3;3;4",
"soundness_avg": 3.3333333333333335,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;3;4",
"presentation_avg": 3.33333333333333... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.953437"
} | {
"id": "AeoOcSFWRg",
"metareview": "The paper proposes an algorithm called Kernel Banzhaf for estimating Banzhaf values, which are an alternative to Shapley values. The algorithm is inspired by KernelSHAP and leverages the connection between Banzhaf values and linear regression.Theoretical analysis and numerical... | {
"decision": "Reject"
} |
7MYu2xO4pp | 2407.17356v1 | Gradient-based inference of abstract task representations for generalization in neural networks | {
"content": "## Abstract\n\nAbstract Humans and many animals show remarkably adaptive behavior and can respond differently to the same input depending on their internal goals. The brain not only represents the intermediate abstractions needed to perform a computation but also actively maintains a representation of t... | [
{
"id": "xbM40qKZZH",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The authors introduce Gradient-Based Inference (GBI) of abstract task representations, a method that enables neural networks to infer and adapt task representatio... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "2;3;3;3",
"confidence_avg": 2.75,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "1;3;2;3",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.954216"
} | {
"id": "GrcWBLXtS1",
"metareview": "This paper proposes to use Gradient-Based Inference (GBI) for learning abstract task representations, which allows neural networks to infer and adapt task representations dynamically. Inspired by human adaptability—where task abstractions allow flexible responses to the same inp... | {
"decision": "Reject"
} |
7NHF4txacw | 2408.05802v1 | Egocentric Vision Language Planning | {
"content": "## Abstract\n\nAbstract We explore leveraging large multi-modal models (LMMs) and text2image models to build a more general embodied agent. LMMs excel in planning long-horizon tasks over symbolic abstractions but struggle with grounding in the physical world, often failing to accurately identify object ... | [
{
"id": "NtskocMB72",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This work has collected a dataset on Virtualhome viewing an action of the agent as a trajectory, with egocentric information. The EgoPlan framework is introduced... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "5;4;4;4",
"confidence_avg": 4.25,
"soundness": "1;2;1;3",
"soundness_avg": 1.75,
"contribution": "1;2;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": "2026-01-14T22:16:00.954910"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7NL74jUiMg | 2410.15748v1 | Alchemy: Amplifying Theorem-Proving Capability Through Symbolic Mutation | {
"content": "## Abstract\n\nAbstract Formal proofs are challenging to write even for experienced experts.\nRecent progress in Neural Theorem Proving (NTP) shows promise in expediting this process.\nHowever, the formal corpora available on the Internet are limited compared to the general text, posing a significant da... | [
{
"id": "ura8LOZ0Nv",
"initial_rating": 6,
"confidence": 5,
"soundness": 4,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a symbolic method called Alchemy to augment formal theorem proving data. Specifically, it mutates \"the candidate theorem by replacing the c... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;5;3",
"confidence_avg": 4,
"soundness": "3;3;4;3",
"soundness_avg": 3.25,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.955583"
} | {
"id": "hjOAboqMGw",
"metareview": "This paper concerns data augmentation for theorem proving in Lean through symbolic rewriting of hypotheseses and proofs. Evaluations performed on Mathlib datasets show that, with data augumentation, further pretraining and finetuning improves the performance by 2.69% on the rand... | {
"decision": "Accept (Poster)"
} |
7PGluppo4k | 2409.13724v1 | Logically Consistent Language Models via Neuro-Symbolic Integration | {
"content": "## Abstract\n\nAbstract Large language models (LLMs) are a promising venue for natural language understanding and generation.\nHowever, current LLMs are far from reliable: they are prone to generating non-factual information and,\nmore crucially, to contradicting themselves when prompted to reason about... | [
{
"id": "YzFlT82Ydc",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper explores improving LLMs' factuality and logical consistency through neuro-symbolic reasoning. It introduces a neuro-symbolic loss function that is used ... | {
"rating": "3;6;6;6;8",
"rating_avg": 5.8,
"confidence": "3;3;3;4;3",
"confidence_avg": 3.2,
"soundness": "2;3;3;3;4",
"soundness_avg": 3,
"contribution": "3;3;3;2;3",
"contribution_avg": 2.8,
"presentation": "2;3;3;3;3",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.956332"
} | {
"id": "6aa8JdhKL1",
"metareview": "The paper proposes LoCo-LMs, a neuro-symbolic fine-tuning method using semantic loss to enhance LLMs’ logical consistency and factuality. It reduces reliance on external tools and shows improved consistency and generalization over baselines. Strengths include novelty, efficiency... | {
"decision": "Accept (Poster)"
} |
7PLpiVdnUC | 2410.02698v1 | Lie Algebra Canonicalization: Equivariant Neural Operators under arbitrary Lie Groups | {
"content": "## Abstract\n\nAbstract The quest for robust and generalizable machine learning models has driven recent interest in exploiting symmetries through equivariant neural networks. In the context of PDE solvers, recent works have shown that Lie point symmetries can be a useful inductive bias for Physics-Info... | [
{
"id": "ESIWmFLGaZ",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 1,
"summary": "The paper proposes a Lie algebra canonicalization mechanism for achieving equivariance with respect to a variety of Lie groups. The authors propose an extension o... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "3;3;2;4",
"confidence_avg": 3,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3;3",
"contribution_avg": 2.5,
"presentation": "1;1;2;3",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.957353"
} | {
"id": "BRDJSzwY45",
"metareview": "This paper proposes a novel extension of canonicalization-like approaches to non-compact Lie groups, a significant advancement in building symmetric neural networks. Frame averaging and canonicalization are foundational techniques, and extending these mechanisms to a broader cla... | {
"decision": "Accept (Poster)"
} |
7PQnFTbizU | 2407.13032v1 | Agent-E: From Autonomous Web Navigation to Foundational Design Principles in Agentic Systems | {
"content": "## Abstract\n\nAbstract AI Agents are changing the way work gets done, both in consumer and enterprise domains. However, the design patterns and architectures to build highly capable agents or multi-agent systems are still developing, and the understanding of the implication of various design choices an... | [
{
"id": "1YB4Er47R6",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 4,
"presentation": 3,
"summary": "The paper introduces Agent-E, a web agent designed to perform complex web-based tasks more efficiently. Agent-E employs a novel hierarchical architecture comprisi... | {
"rating": "5;5;5",
"rating_avg": 5,
"confidence": "3;4;5",
"confidence_avg": 4,
"soundness": "2;3;3",
"soundness_avg": 2.6666666666666665,
"contribution": "3;2;4",
"contribution_avg": 3,
"presentation": "3;2;3",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.958408"
} | {
"id": "jTc267WAiu",
"metareview": "This paper introduces Agent-E, a hierarchical LLM-powered web agent with innovative mechanisms like flexible DOM distillation and self-refinement. While the authors demonstrate promising results on the WebVoyager benchmark, a key weakness lies in the limited evaluation scope. De... | {
"decision": "Reject"
} |
7QGyDi9VsO | 2410.04940v1 | Next state prediction gives rise to entangled, yet compositional representations of objects | {
"content": "## Abstract\n\nAbstract Compositional representations are thought to enable humans to generalize across combinatorially vast state spaces. Models with learnable object slots, which encode information about objects in separate latent codes, have shown promise for this type of generalization but rely on s... | [
{
"id": "LCckUdkk4f",
"initial_rating": 6,
"confidence": 2,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The paper studies object disentanglement in the representation of visual scenes. Specifically, object-centric representations enforce an inductive bias for disent... | {
"rating": "1;5;6;6",
"rating_avg": 4.5,
"confidence": "5;4;3;2",
"confidence_avg": 3.5,
"soundness": "1;2;2;3",
"soundness_avg": 2,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.959023"
} | {
"id": "wZQHeJ0h7u",
"metareview": "This paper studies structured (object-centric) vs. unstructured visual representations in the context of next-state prediction. While the paper is well written and tackles a very interesting problem, the reviewers voiced mixed opinions about the strength of the experimental eval... | {
"decision": "Reject"
} |
7Qa2SpjxIS | 2406.07358v3 | AI Sandbagging: Language Models can Strategically Underperform on Evaluations | {
"content": "## Abstract\n\nAbstract Trustworthy capability evaluations are crucial for ensuring the safety of AI systems, and are becoming a key component of AI regulation. However, the developers of an AI system, or the AI system itself, may have incentives for evaluations to understate the AI’s actual capability.... | [
{
"id": "AdmeXuyziM",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes the concept of sandbagging—a phenomenon where AI models are intentionally fine-tuned or prompted to underperform in certain evaluations. The p... | {
"rating": "3;5;6",
"rating_avg": 4.666666666666667,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "2;2;2",
"soundness_avg": 2,
"contribution": "2;3;3",
"contribution_avg": 2.6666666666666665,
"presentation": "1;3;3",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.959818"
} | {
"id": "Y4mOqADfXb",
"metareview": "This paper proposes the concept of sandbagging—a phenomenon where AI models are intentionally fine-tuned or prompted to underperform in certain evaluations. The paper demonstrates that large language models, like GPT-4 and Claude 3, can be conditioned to underperform on dangerou... | {
"decision": "Accept (Poster)"
} |
7RVJxmtzTj | 2403.06403v4 | PointSeg: A Training-Free Paradigm for 3D Scene Segmentation via Foundation Models | {
"content": "## Abstract\n\nAbstract Recent success of vision foundation models have shown promising performance for the 2D perception tasks. However, it is difficult to train a 3D foundation network directly due to the limited dataset and it remains under explored whether existing foundation models can be lifted to... | [
{
"id": "vj9AMDNusC",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "This paper introduces PointSeg, a training-free framework designed to leverage existing 2D vision foundation models for 3D scene segmentation tasks. PointSeg achi... | {
"rating": "5;5;5;6",
"rating_avg": 5.25,
"confidence": "3;3;3;3",
"confidence_avg": 3,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"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": "2026-01-14T22:16:00.960643"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7S1xDos9pH | 2408.02295v2 | Generalized Gaussian Temporal Difference Error for Uncertainty-aware Reinforcement Learning | {
"content": "## Abstract\n\nAbstract Conventional uncertainty-aware temporal difference (TD) learning methods often rely on simplistic assumptions, typically including a zero-mean Gaussian distribution for TD errors. Such oversimplification can lead to inaccurate error representations and compromised uncertainty est... | [
{
"id": "mB4R5FgZmB",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The authors consider uncertainty estimation in TD learning and seek to generalize the conceptually simple and implicit Gaussian assumption behind the MSE loss. To... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;3;3",
"confidence_avg": 3.25,
"soundness": "3;3;3;3",
"soundness_avg": 3,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "1;1;3;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.961379"
} | {
"id": "YH5y61FnW1",
"metareview": "This paper addresses the limitations of conventional uncertainty-aware Temporal Difference (TD) learning methods, and proposes a more sophisticated and flexible way to model uncertainty in TD learning by using the GGD and explicitly considering higher-order moments, leading to i... | {
"decision": "Reject"
} |
7SFTZwNUQA | 2410.11730v1 | Patch-Based Diffusion Models Beat Whole-Image Models for Mismatched Distribution Inverse Problems | {
"content": "## Abstract\n\nAbstract Diffusion models have achieved excellent success in solving inverse problems\ndue to their ability to learn strong image priors,\nbut existing approaches require a large training dataset of images\nthat should come from the same distribution as the test dataset.\nWhen the trainin... | [
{
"id": "PEOiFy8qrI",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 2,
"summary": "The paper considers the problem of adapting diffusion models trained on a domain A to the task of solving reconstruction problems on another domain B, and invest... | {
"rating": "3;5;5;5;6",
"rating_avg": 4.8,
"confidence": "5;4;4;3;4",
"confidence_avg": 4,
"soundness": "3;2;2;3;3",
"soundness_avg": 2.6,
"contribution": "2;3;2;2;2",
"contribution_avg": 2.2,
"presentation": "2;1;1;2;3",
"presentation_avg": 1.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.962133"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7UgQjFEadn | 2407.03604v1 | Modality-Specialized Synergizers for Interleaved Vision-Language Generalists | {
"content": "## Abstract\n\nAbstract Recent advancements in Vision-Language Models (VLMs) have led to the development of Vision-Language Generalists (VLGs) capable of understanding and generating interleaved images and text.\nDespite these advances, VLGs still struggle to follow user instructions for interleaved tex... | [
{
"id": "ntVG9J8dJK",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper aims to improve the understanding ability of existing Vision-Language Generalists (VLGs) and propose the Modality-Specialized Synergizers (MoSS). Moss ... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "3;4;3;3",
"confidence_avg": 3.25,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "2;3;3;3",
"contribution_avg": 2.75,
"presentation": "2;3;3;2",
"presentation_avg": 2.5
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.962838"
} | {
"id": "3QRNfCZekO",
"metareview": "This paper introduces Modality-Specialized Synergizers to improve interleaved text-image generation in Vision-Language Generalists (VLGs). By integrating modality-specific LoRAs and providing a large-scale interleaved instruction-tuning dataset (LeafInstruct), the authors show m... | {
"decision": "Accept (Poster)"
} |
7UqQJUKaLM | 2405.11874v2 | xFinder: Large Language Models as Automated Evaluators for Reliable Evaluation | {
"content": "## Abstract\n\nAbstract The continuous advancement of large language models (LLMs) has brought increasing attention to the critical issue of developing fair and reliable methods for evaluating their performance. Particularly, the emergence of subjective or non-subjective cheating phenomena, such as test... | [
{
"id": "RXMkvtQnDN",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper proposes a novel evaluator for answer extraction and matching in LLM evaluation. The main idea is to first construct a large-scale LLM response evaluat... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "5;4;4;3",
"confidence_avg": 4,
"soundness": "4;2;3;3",
"soundness_avg": 3,
"contribution": "4;2;3;3",
"contribution_avg": 3,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.963456"
} | {
"id": "tOjptij52Z",
"metareview": "The paper proposed to improve the robustness of answer extraction when evaluating large language models, focusing on the limitations of existing RegEx-based evaluation frameworks. It argues that these methods often lead to extraction errors and unreliable evaluations. The paper ... | {
"decision": "Accept (Poster)"
} |
7X3fi8aJBL | 2409.11598v1 | Towards Fair RAG: On the Impact of Fair Ranking in Retrieval-Augmented Generation | {
"content": "## Abstract\n\nAbstract Many language models now enhance their responses with retrieval capabilities, leading to the widespread adoption of retrieval-augmented generation (RAG) systems. However, despite retrieval being a core component of RAG, much of the research in this area overlooks the extensive bo... | [
{
"id": "mnwqXcu9K7",
"initial_rating": 8,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "In this paper, the authors investigate the impact of fair ranking on RAG systems. They conduct systematic evaluations of RAG systems integrated with fair ranking... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "3;2;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.964131"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7X65yoKl3Y | 2410.09692v1 | ALLoRA: Adaptive Learning Rate Mitigates LoRA Fatal Flaws | {
"content": "## Abstract\n\nAbstract Low-Rank Adaptation (LoRA) is the bread and butter of Large Language Model (LLM) finetuning. LoRA learns an additive low-rank perturbation, 𝑨 𝑩 𝑨 𝑩 {\\bm{A}}{\\bm{B}} bold_italic_A bold_italic_B , of a pretrained matrix parameter 𝑾 𝑾 {\\bm{W}} bold_italic_W to align the m... | [
{
"id": "GRD791fcy0",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "This paper first identifies three main challenges in fine-tuning with LoRA: dropout, zero initialization, and the difficulty of setting an appropriate scaling fac... | {
"rating": "3;3;3;3;3;5",
"rating_avg": 3.3333333333333335,
"confidence": "4;4;4;4;4;3",
"confidence_avg": 3.8333333333333335,
"soundness": "2;2;2;2;2;2",
"soundness_avg": 2,
"contribution": "2;1;2;2;3;2",
"contribution_avg": 2,
"presentation": "2;2;3;1;3;2",
"presentation_avg": 2.1666666666666665
... | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.964875"
} | {
"id": "2dTfTIXMhd",
"metareview": "This work first analyzed the limitations of the vanilla LoRA. In total, the vanilla LoRA has three weak points, some of which were already known, e.g., the scaling effect of LoRA. In addition, reviewers pointed out (potentially) confusing parts in their proofs and analyses. Howe... | {
"decision": "Reject"
} |
7XIkRgYjK3 | 2410.08893v1 | Drama: Mamba-Enabled Model-Based Reinforcement Learning Is Sample and Parameter Efficient | {
"content": "## Abstract\n\nAbstract Model-based reinforcement learning (RL) offers a solution to the data inefficiency that plagues most model-free RL algorithms. However, learning a robust world model often demands complex and deep architectures, which are expensive to compute and train. Within the world model, dy... | [
{
"id": "Thwrl17mCr",
"initial_rating": 5,
"confidence": 4,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces DRAMA, a model-based reinforcement learning (MB-RL) agent that leverages the Mamba-2 architecture, a state space model (SSM), as its core dy... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "3;4;4;4",
"confidence_avg": 3.75,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "3;3;2;3",
"contribution_avg": 2.75,
"presentation": "1;3;1;3",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.965593"
} | {
"id": "9ua4eAHbcJ",
"metareview": "The paper presents a significant contribution by effectively incorporating Mamba architecture into model-based RL, achieving competitive performance with only 7M parameters. While reviewers raised concerns about long-term dependency claims and computational efficiency, the autho... | {
"decision": "Accept (Poster)"
} |
7XNgVPxCiA | 2410.01322v1 | Forte : Finding Outliers with Representation Typicality Estimation | {
"content": "## Abstract\n\nAbstract Generative models can now produce photorealistic synthetic data which is virtually indistinguishable from the real data used to train it. This is a significant evolution over previous models which could produce reasonable facsimiles of the training data, but ones which could be v... | [
{
"id": "uI15epqkvY",
"initial_rating": 10,
"confidence": 4,
"soundness": 4,
"contribution": 4,
"presentation": 4,
"summary": "The authors present a methodology, Forte, which enables the identification of outliers using a rigorous unsupervised algorithm. The algorithm relies on establis... | {
"rating": "3;5;6;10",
"rating_avg": 6,
"confidence": "4;2;2;4",
"confidence_avg": 3,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "2;2;2;4",
"contribution_avg": 2.5,
"presentation": "1;1;3;4",
"presentation_avg": 2.25
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.966625"
} | {
"id": "xzLVEwgco7",
"metareview": "The paper focuses on outlier detection, and is well received by all reviewers. There were concerns regarding the clarity and novelty, however, these were well addressed, even raising scores significantly during the rebuttal phase. Thus, I recommend acceptance.",
"additional_co... | {
"decision": "Accept (Poster)"
} |
7XgKAabsPp | 2406.16437v2 | Theory on Mixture-of-Experts in Continual Learning | {
"content": "## Abstract\n\nAbstract Continual learning (CL) has garnered significant attention because of its ability to adapt to new tasks that arrive over time. Catastrophic forgetting (of old tasks) has been identified as a major issue in CL, as the model adapts to new tasks.\nThe Mixture-of-Experts (MoE) model ... | [
{
"id": "uIQd1ahlJu",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "This paper provides a theoretical study of Mixture-of-Experts (MoE) models for Continual Learning (CL). Specifically, it examines the CL of linear regression task... | {
"rating": "5;5;8",
"rating_avg": 6,
"confidence": "4;4;3",
"confidence_avg": 3.6666666666666665,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "2;2;3",
"contribution_avg": 2.3333333333333335,
"presentation": "3;3;2",
"presentation_avg": 2.6666666666666665
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Spotlight",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.967800"
} | {
"id": "R6zXjB6LZl",
"metareview": "This paper provides a theoretical study of Mixture-of-Experts (MoE) models for Continual Learning (CL). Specifically, it examines the CL of linear regression tasks in an over-parameterized regime. The benefit of this regime is that each task has multiple possible solutions, incr... | {
"decision": "Accept (Spotlight)"
} |
7ZeoPg3eTA | 2403.15879v6 | TrustSQL: Benchmarking Text-to-SQL Reliability with Penalty-Based Scoring | {
"content": "## Abstract\n\nAbstract Text-to-SQL enables users to interact with databases using natural language, simplifying the retrieval and synthesis of information.\nDespite the remarkable success of large language models (LLMs) in translating natural language questions into SQL queries, widespread deployment r... | [
{
"id": "CBGWoVLAFd",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 3,
"presentation": 2,
"summary": "The paper introduces TrustSQL, a benchmark designed to assess the reliability of text-to-SQL systems in handling both feasible and infeasible questions. To build ... | {
"rating": "3;5;5;5",
"rating_avg": 4.5,
"confidence": "5;4;3;4",
"confidence_avg": 4,
"soundness": "3;3;3;2",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "1;3;1;2",
"presentation_avg": 1.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.968923"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7Zppme1swQ | 2403.02877v1 | ActiveAD: Planning-Oriented Active Learning for End-to-End Autonomous Driving | {
"content": "## Abstract\n\nAbstract End-to-end differentiable learning for autonomous driving (AD) has recently become a prominent paradigm. One main bottleneck lies in its voracious appetite for high-quality labeled data e.g. 3D bounding boxes and semantic segmentation, which are notoriously expensive to manually ... | [
{
"id": "n2U9Uk7aux",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper proposes ActiveAD, an active learning framework for end-to-end autonomous driving. One major challenge in E2E-AD lies in the expensive data annotation p... | {
"rating": "3;3;3;6",
"rating_avg": 3.75,
"confidence": "5;5;4;4",
"confidence_avg": 4.5,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;2;3;4",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.969607"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7dmsy2Vd5h | 2407.14129v2 | Comparing and Contrasting Deep Learning Weather Prediction Backbones on Navier-Stokes and Atmospheric Dynamics | {
"content": "## Abstract\n\nAbstract Remarkable progress in the development of Deep Learning Weather Prediction (DLWP) models positions them to become competitive with traditional numerical weather prediction (NWP) models.\nIndeed, a wide number of DLWP architectures—based on various backbones, including U-Net, Tran... | [
{
"id": "HbXbaFmQYi",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "The paper aims to conduct a comprehensive evaluation and comparisons of deep learning backbones for weather forecasting. Authors selected seven widely used networ... | {
"rating": "3;3;5;8",
"rating_avg": 4.75,
"confidence": "4;5;4;3",
"confidence_avg": 4,
"soundness": "2;2;3;4",
"soundness_avg": 2.75,
"contribution": "2;2;2;3",
"contribution_avg": 2.25,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.970919"
} | {
"id": "AdGt0lPJJd",
"metareview": "This paper is primarily experimental and compares widely used deep learning models for weather prediction.\n\nThere is one review in strong support, while the rest are critical of the work. The biggest criticisms center around lack of innovation (clarity of contribution), experi... | {
"decision": "Reject"
} |
7egJb0X9m2 | 2210.15050v2 | TILDE-Q: a Transformation Invariant Loss Function for Time-Series Forecasting | {
"content": "## Abstract\n\nAbstract Time-series forecasting has gained increasing attention in the field of artificial intelligence due to its potential to address real-world problems across various domains, including energy, weather, traffic, and economy. While time-series forecasting is a well-researched field, p... | [
{
"id": "WTs7qOF52l",
"initial_rating": 5,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces a novel loss function for time-series forecasting. Traditional loss functions like Mean Squared Error (MSE) or Dynamic Time Warping (DTW) ar... | {
"rating": "3;5;6;6",
"rating_avg": 5,
"confidence": "4;5;4;3",
"confidence_avg": 4,
"soundness": "2;3;3;3",
"soundness_avg": 2.75,
"contribution": "3;3;3;2",
"contribution_avg": 2.75,
"presentation": "4;3;3;3",
"presentation_avg": 3.25
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.971709"
} | {
"id": "AqawDOe1ef",
"metareview": "This paper introduces TILDE-Q, a novel shape-aware loss function for time-series forecasting that captures amplitude and phase distortions while modeling both periodic and non-periodic dynamics. By addressing the limitations of traditional loss functions like MSE and MAE, TILDE-... | {
"decision": "Reject"
} |
7gGVDrqVaz | 2410.11133v1 | 3D-Prover: Diversity Driven Theorem Proving With Determinantal Point Processes | {
"content": "## Abstract\n\nAbstract A key challenge in automated formal reasoning is the\nintractable search space, which grows exponentially with the depth of the proof.\nThis branching is caused by\nthe large number of candidate proof tactics which can be applied to a given goal.\nNonetheless, many of these tacti... | [
{
"id": "lsLBzFTpjq",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "The search space of proofs grows exponentially with the depth of the proof but a number of branches in this space capture sequences that could be semantically sim... | {
"rating": "5;5;6;6",
"rating_avg": 5.5,
"confidence": "4;4;4;3",
"confidence_avg": 3.75,
"soundness": "2;3;4;3",
"soundness_avg": 3,
"contribution": "3;2;3;3",
"contribution_avg": 2.75,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.973041"
} | {
"id": "FPHbeTbaFs",
"metareview": "This paper concerns pruning proof search of neural theorem proving and proposes 3D-Prover, which is built on top of the previous work ReProver and consists of a representation learning component capturing the transition semantic of theorem proving, and a filtering mechansim for ... | {
"decision": "Reject"
} |
7hM5597bCv | 2410.11338v1 | DIAR: Diffusion-model-guided Implicit Q-learning with Adaptive Revaluation | {
"content": "## Abstract\n\nAbstract We propose a novel offline reinforcement learning (offline RL) approach, introducing the Diffusion-model-guided Implicit Q-learning with Adaptive Revaluation (DIAR) framework. We address two key challenges in offline RL: out-of-distribution samples and long-horizon problems. We l... | [
{
"id": "MHKgrRQNfe",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 1,
"presentation": 1,
"summary": "The paper proposes an offline RL algorithm utilizing latent diffusion skill models for temporal abstraction, and Q learning with these skills. During policy rollo... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "3;5;4;4",
"confidence_avg": 4,
"soundness": "2;2;2;2",
"soundness_avg": 2,
"contribution": "2;2;1;2",
"contribution_avg": 1.75,
"presentation": "3;2;1;2",
"presentation_avg": 2
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.973665"
} | {
"id": "fTea47CCrQ",
"metareview": "Authors present a model-based offline RL method using diffusion models that outputs sequence-level distributions to handle long horizons and deal with compounding error issues that occur with 1-step models. Their method also addresses out-of-distribution samples with a learned v... | {
"decision": "Reject"
} |
7ha61H73pg | 2410.17875v1 | Understanding Layer Significance in LLM Alignment | {
"content": "## Abstract\n\nAbstract Aligning large language models (LLMs) through fine-tuning is essential for tailoring them to specific applications. Therefore, understanding what LLMs learn during the alignment process is crucial. Recent studies suggest that alignment primarily adjusts a model’s presentation sty... | [
{
"id": "hxzUKFGiPq",
"initial_rating": 6,
"confidence": 4,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "This paper introduces a novel and interesting approach, Important Layers for Alignment (ILA), to enhance the fine-tuning efficiency of large language models (LLM)... | {
"rating": "3;3;3;5;6",
"rating_avg": 4,
"confidence": "5;4;4;3;4",
"confidence_avg": 4,
"soundness": "2;3;1;3;3",
"soundness_avg": 2.4,
"contribution": "2;1;2;2;2",
"contribution_avg": 1.8,
"presentation": "4;2;1;2;3",
"presentation_avg": 2.4
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.974328"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7heZQqlY5t | 2410.04560v1 | GAMformer: In-Context Learning for Generalized Additive Models | {
"content": "## Abstract\n\nAbstract Generalized Additive Models (GAMs) are widely recognized for their ability to create fully interpretable machine learning models for tabular data. Traditionally, training GAMs involves iterative learning algorithms, such as splines, boosted trees, or neural networks, which refine... | [
{
"id": "lmBB2p36Zc",
"initial_rating": 5,
"confidence": 3,
"soundness": 3,
"contribution": 2,
"presentation": 3,
"summary": "The paper presents GAMformer, a novel model for fitting Generalized Additive Models (GAMs) using in-context learning (ICL) within a transformer-based framework. U... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "5;3;3;3",
"confidence_avg": 3.5,
"soundness": "2;2;3;3",
"soundness_avg": 2.5,
"contribution": "1;2;2;3",
"contribution_avg": 2,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.975046"
} | {
"id": "VSHMhwCQlz",
"metareview": "The paper introduces GAMformer, a method that uses in-context learning to fit Generalized Additive Models in a single step, rather than traditional iterative methods. The models presented are trained on synthetic data but demonstrates good performance on real-world datasets. The... | {
"decision": "Reject"
} |
7igPXQFupX | 2310.10845v2 | CoTFormer: A Chain of Thought Driven Architecture with Budget-Adaptive Computation Cost at Inference | {
"content": "## Abstract\n\nAbstract Scaling language models to larger and deeper sizes has led to significant boosts in performance. Even though the size of these models limits their application in compute-constrained environments, the race to continually develop ever larger and deeper foundational models is underw... | [
{
"id": "cEeV0gUERe",
"initial_rating": 6,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 3,
"summary": "This paper introduces CoTFormer, a novel transformer architecture that draws inspiration from chain-of-thought (CoT) reasoning. The key insight is recognizing tha... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "3;3;4;3",
"confidence_avg": 3.25,
"soundness": "2;2;2;3",
"soundness_avg": 2.25,
"contribution": "2;3;2;3",
"contribution_avg": 2.5,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.975675"
} | {
"id": "kjheq4zbUH",
"metareview": "Summary:\nThe paper introduces CoTFormer, a novel transformer architecture that draws inspiration from chain-of-thought reasoning. The key innovation is allowing tokens to attend to representations from all previous \"thought\" steps, which differs from both standard transformer... | {
"decision": "Accept (Poster)"
} |
7liN6uHAQZ | 2311.01806v1 | Sketching for Convex and Nonconvex Regularized Least Squares with Sharp Guarantees | {
"content": "## Abstract\n\nAbstract Randomized algorithms are important for solving large-scale optimization problems. In this paper, we propose a fast sketching algorithm for least square problems regularized by convex or nonconvex regularization functions, Sketching for Regularized Optimization (SRO). Our SRO alg... | [
{
"id": "uWGMdYKmzm",
"initial_rating": 8,
"confidence": 3,
"soundness": 3,
"contribution": 3,
"presentation": 2,
"summary": "The paper introduces an extension of the Iterative Hessian Sketch from Pilanci and Wainwright to the regularized setting. The authors first show convergence of th... | {
"rating": "3;5;5",
"rating_avg": 4.333333333333333,
"confidence": "4;3;4",
"confidence_avg": 3.6666666666666665,
"soundness": "3;3;3",
"soundness_avg": 3,
"contribution": "1;2;3",
"contribution_avg": 2,
"presentation": "3;3;1",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.976623"
} | {
"id": "VnrmBX6Icz",
"metareview": "Dear Authors,\n\nThank you for your valuable contribution to the ICLR and the ML community. Your submitted paper has undergone a rigorous review process, and I have carefully read and considered the feedback provided by the reviewers.\n\nThis paper introduces an extension of the... | {
"decision": "Accept (Poster)"
} |
7o6SG5gVev | 2410.00752v1 | TestGenEval: A Real World Unit Test Generation and Test Completion Benchmark | {
"content": "## Abstract\n\nAbstract Code generation models can help improve many common software tasks ranging from code completion to defect prediction. Most of the existing benchmarks for code generation LLMs focus on code authoring or code completion.\nSurprisingly, there has been far less effort dedicated to be... | [
{
"id": "Ink4zyU3nR",
"initial_rating": 8,
"confidence": 5,
"soundness": 3,
"contribution": 3,
"presentation": 4,
"summary": "This paper introduces a new LLM for test generation benchmark, TestGenEval. It builds on the same repositories as SWEBench, but focuses on extracting the test fil... | {
"rating": "5;5;6;8",
"rating_avg": 6,
"confidence": "4;4;4;5",
"confidence_avg": 4.25,
"soundness": "3;2;3;3",
"soundness_avg": 2.75,
"contribution": "3;2;4;3",
"contribution_avg": 3,
"presentation": "2;3;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.977974"
} | {
"id": "xFEcgWnQnX",
"metareview": "This paper introduces a new benchmark to support the research on code generation using large language models (LLMs) for real-world software engineering applications. Specifically, the benchmark covers tasks including producing initial test cases, test suite completion, and code ... | {
"decision": "Accept (Poster)"
} |
7oaWthT9EO | 2405.16351v1 | A Differential Equation Approach for Wasserstein GANs and Beyond | {
"content": "## Abstract\n\nAbstract We propose a new theoretical lens to view Wasserstein generative adversarial networks (WGANs). In our framework, we define a discretization inspired by a distribution-dependent ordinary differential equation (ODE). We show that such a discretization is convergent and propose a vi... | [
{
"id": "ulLhkMz0dJ",
"initial_rating": 6,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 1,
"summary": "This paper presents a new variant of Wasserstein generative adversarial networks (WGANs) based on a distribution-dependent ordinary differential equation (ODE). I... | {
"rating": "5;6;6",
"rating_avg": 5.666666666666667,
"confidence": "4;2;3",
"confidence_avg": 3,
"soundness": "2;3;2",
"soundness_avg": 2.3333333333333335,
"contribution": "2;3;2",
"contribution_avg": 2.3333333333333335,
"presentation": "3;3;1",
"presentation_avg": 2.3333333333333335
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.979058"
} | {
"id": "AS1g3s9YR5",
"metareview": "This paper is concerned with the Wasserstein generative adversarial network (WGAN). It introduces a notation of gradient flow associated with WGAN leveraging the linear functional derivative. Based on it, the authors propose an algorithm to train WGAN. The major criticism is on ... | {
"decision": "Reject"
} |
7rxn2wnx88 | 2411.05830v1 | Unmasking the Version-Switching Capabilities of Code Generation Models | {
"content": "## Abstract\n\nAbstract The rapid evolution of software libraries presents a significant challenge for code generation models, which must adapt to frequent version updates while maintaining compatibility with previous versions. Existing code completion benchmarks often overlook this dynamic aspect, and ... | [
{
"id": "aDofdCBuav",
"initial_rating": 3,
"confidence": 5,
"soundness": 2,
"contribution": 1,
"presentation": 3,
"summary": "The paper argues that LLMs are extensively used as virtual coding assistants but frequently fail to generate accurate version-specific code amidst the fast-evolvi... | {
"rating": "3;3;3;5",
"rating_avg": 3.5,
"confidence": "4;4;5;4",
"confidence_avg": 4.25,
"soundness": "3;3;2;2",
"soundness_avg": 2.5,
"contribution": "2;2;1;2",
"contribution_avg": 1.75,
"presentation": "3;3;3;3",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.979703"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7tOc6h8bea | 2410.02725v1 | Adaptive Inference-Time Compute: LLMs Can Predict if They Can Do Better, Even Mid-Generation | {
"content": "## Abstract\n\nAbstract Inference-time computation is a powerful paradigm to enhance the performance of large language models (LLMs), with Best-of-N sampling being a widely used technique. However, this method is computationally expensive, requiring both (1) an external reward model and (2) the generati... | [
{
"id": "EU19LM665b",
"initial_rating": 5,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "- The paper proposes to simple way to train a LLM to do self-evaluation (with additional tokens) and uses the logits for the trained additional tokens to decide w... | {
"rating": "3;5;5;6",
"rating_avg": 4.75,
"confidence": "4;3;4;4",
"confidence_avg": 3.75,
"soundness": "2;3;2;3",
"soundness_avg": 2.5,
"contribution": "2;2;2;2",
"contribution_avg": 2,
"presentation": "2;3;3;3",
"presentation_avg": 2.75
} | {
"primary_area": "",
"track": "main",
"venue": "Submitted to ICLR 2025",
"venueid": "ICLR.cc/2025/Conference/Rejected_Submission",
"processed_at": "2026-01-14T22:16:00.980343"
} | {
"id": "iTR8GznsfM",
"metareview": "In this paper, the authors propose a new strategy for efficient inference-time improvement of LLMs. Specifically, the authors exploit adaptive sampling (selecting the N in best-of-N as a function of the prompt) + early pruning (pruning unpromising mid-generations) upon the learn... | {
"decision": "Reject"
} |
7vH8DO2oPk | 2407.01607v1 | Multi-Epoch Learning with Data Augmentation for Deep Click-Through Rate Prediction | {
"content": "## Abstract\n\nAbstract. This paper investigates the one-epoch overfitting phenomenon in Click-Through Rate (CTR) models, where performance notably declines at the start of the second epoch. Despite extensive research, the efficacy of multi-epoch training over the conventional one-epoch approach remains... | [
{
"id": "DaqavQgr8e",
"initial_rating": 3,
"confidence": 4,
"soundness": 2,
"contribution": 2,
"presentation": 3,
"summary": "This paper investigates OEO, considering overfitting in the embedding layer caused by high-dimensional data sparsity as a critical issue. To address this, it prop... | {
"rating": "3;3;3;3;6",
"rating_avg": 3.6,
"confidence": "4;4;5;4;4",
"confidence_avg": 4.2,
"soundness": "3;3;3;2;4",
"soundness_avg": 3,
"contribution": "2;2;2;2;4",
"contribution_avg": 2.4,
"presentation": "1;3;3;3;4",
"presentation_avg": 2.8
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Conference Withdrawn Submission",
"venueid": "ICLR.cc/2025/Conference/Withdrawn_Submission",
"processed_at": "2026-01-14T22:16:00.980983"
} | {
"id": "",
"metareview": "",
"additional_comments": ""
} | {
"decision": ""
} |
7visV100Ms | 2410.06961v1 | Self-Boosting Large Language Models with Synthetic Preference Data | {
"content": "## Abstract\n\nAbstract Through alignment with human preferences, Large Language Models (LLMs) have advanced significantly in generating honest, harmless, and helpful responses.\nHowever, collecting high-quality preference data is a resource-intensive and creativity-demanding process, especially for the... | [
{
"id": "QNODZoBq4m",
"initial_rating": 5,
"confidence": 3,
"soundness": 2,
"contribution": 2,
"presentation": 2,
"summary": "The paper presents \"SynPO\", which trains a self-prompt generator by controlling the keywords used. Noises (extracted also from responses) are inserted to produc... | {
"rating": "5;5;6;8;8",
"rating_avg": 6.4,
"confidence": "3;3;4;2;4",
"confidence_avg": 3.2,
"soundness": "3;2;3;3;3",
"soundness_avg": 2.8,
"contribution": "2;2;3;3;3",
"contribution_avg": 2.6,
"presentation": "2;2;4;3;4",
"presentation_avg": 3
} | {
"primary_area": "",
"track": "main",
"venue": "ICLR 2025 Poster",
"venueid": "ICLR.cc/2025/Conference",
"processed_at": "2026-01-14T22:16:00.981641"
} | {
"id": "gOnJ2Iz2qB",
"metareview": "This paper introduces SynPO, a self-boosting framework that leverages synthetic preference data to iteratively improve the performance of large language models (LLMs). The approach trains a self-prompt generator and response improver to produce synthetic prompts and refine respo... | {
"decision": "Accept (Poster)"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.