Datasets:
paper_id stringlengths 10 10 | year int64 2.03k 2.03k | title stringlengths 12 165 | abstract stringlengths 446 3.32k | authors stringlengths 6 1.02k | author_ids stringlengths 8 1.13k | keywords listlengths 1 28 | primary_area stringclasses 21
values | arxiv_id stringlengths 0 10 | citations_serper int64 1 3.97k ⌀ | decision stringclasses 6
values | decision_binary stringclasses 2
values | reviewer_scores listlengths 1 8 | num_reviewers int64 1 8 | reviews stringlengths 0 55.5k | metareview stringlengths 0 5.01k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SjufxrSOYd | 2,025 | Higher-Order Graphon Neural Networks: Approximation and Cut Distance | Graph limit models, like *graphons* for limits of dense graphs, have recently been used to study size transferability of graph neural networks (GNNs). While most literature focuses on message passing GNNs (MPNNs), in this work we attend to the more powerful *higher-order* GNNs. First, we extend the $k$-WL test for grap... | Daniel Herbst, Stefanie Jegelka | ~Daniel_Herbst1, ~Stefanie_Jegelka3 | [
"graph neural networks",
"invariant graph networks",
"universal approximation",
"graph limits",
"graphons",
"transferability",
"homomorphism densities",
"machine learning theory."
] | learning on graphs and other geometries & topologies | 2503.14338 | 7 | accept (spotlight) | accept | [
8,
8,
8,
8,
8
] | 5 | <thread>
creation date:
1730077724713
thread_type:
official_review
<thread_object>
creator: reviewer_p53n
<content>
summary:
This paper proposes the Invariant Graphon Network (IWN) architecture for graphons as an extension of the Invariant Graph Network GNN architecture for discrete graphs. The paper's results are f... | metareview:
The authors propose Invariant Graphon Networks (IWNs) to generalized Invariant Graph Networks (IGNs) to graphons. The authors prove that IWNs of order (k+1) is at least as powerful as k-dimensional Weisfeiler-Leman (WL) test for graphons. The derived expressivity extend results in Cai and Wang (2022). Addit... |
fU8H4lzkIm | 2,025 | PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems | Solving partial differential equations (PDEs) serves as a cornerstone for modeling complex dynamical systems. Recent progresses have demonstrated grand benefits of data-driven neural-based models for predicting spatiotemporal dynamics (e.g., tremendous speedup gain compared with classical numerical methods). However, m... | Bocheng Zeng, Qi Wang, Mengtao Yan, Yang Liu, Ruizhi Chengze, Yi Zhang, Hongsheng Liu, Zidong Wang, Hao Sun | ~Bocheng_Zeng1, ~Qi_Wang30, ~Mengtao_Yan1, ~Yang_Liu52, ~Ruizhi_Chengze1, ~Yi_Zhang92, ~Hongsheng_Liu1, ~Zidong_Wang1, ~Hao_Sun4 | [
"physics-encoded; spatiotemporal pdes; graph network; deep learning;"
] | applications to physical sciences (physics, chemistry, biology, etc.) | 2410.01337 | 17 | accept (spotlight) | accept | [
8,
8,
6,
10,
8
] | 5 | <thread>
creation date:
1729329119859
thread_type:
official_review
<thread_object>
creator: reviewer_zatm
<content>
summary:
The paper introduces a graph learning approach called Physics-encoded Message Passing Graph Network (PhyMPGN) designed to model spatiotemporal PDE systems on coarse unstructured meshes using s... | metareview:
The paper introduces PhyMPGN, a GNN-based approach for emulating solutions to PDEs on coarse, unstructured meshes. PhyMPGN integrates a learnable Laplace-Beltrami operator and a specialized boundary condition padding strategy at both physical and latent-space level. This is then coupled with ODE solvers for... |
3b9SKkRAKw | 2,025 | LeFusion: Controllable Pathology Synthesis via Lesion-Focused Diffusion Models | Patient data from real-world clinical practice often suffers from data scarcity and long-tail imbalances, leading to biased outcomes or algorithmic unfairness. This study addresses these challenges by generating lesion-containing image-segmentation pairs from lesion-free images. Previous efforts in medical imaging synt... | Hantao Zhang, Yuhe Liu, Jiancheng Yang, Shouhong Wan, Xinyuan Wang, Wei Peng, Pascal Fua | ~Hantao_Zhang2, ~Yuhe_Liu3, ~Jiancheng_Yang3, ~Shouhong_Wan1, ~Xinyuan_Wang2, ~Wei_Peng4, ~Pascal_Fua1 | [
"data synthesis",
"diffusion models",
"cardiac mri",
"lung nodule ct",
"segmentation"
] | applications to physical sciences (physics, chemistry, biology, etc.) | 2403.14066 | 12 | accept (spotlight) | accept | [
8,
8,
8,
8
] | 4 | <thread>
creation date:
1730106179437
thread_type:
official_review
<thread_object>
creator: reviewer_f4py
<content>
summary:
This paper focuses on generating lesion-containing images from healthy images to address challenges in downstream segmentation tasks, such as real-world data scarcity and long-tail distributio... | metareview:
This paper proposes LeFusion, a lesionfocused diffusion model. By redesigning the diffusion learning objectives to focus on lesion areas, the authors simplify the learning process while preserving high-fidelity backgrounds by integrating forward diffused background contexts into the reverse diffusion proces... |
Nx4PMtJ1ER | 2,025 | Signature Kernel Conditional Independence Tests in Causal Discovery for Stochastic Processes | Inferring the causal structure underlying stochastic dynamical systems from observational data holds great promise in domains ranging from science and health to finance. Such processes can often be accurately modeled via stochastic differential equations (SDEs), which naturally imply causal relationships via `which var... | Georg Manten, Cecilia Casolo, Emilio Ferrucci, Søren Wengel Mogensen, Cristopher Salvi, Niki Kilbertus | ~Georg_Manten1, ~Cecilia_Casolo1, ~Emilio_Ferrucci1, ~Søren_Wengel_Mogensen1, ~Cristopher_Salvi1, ~Niki_Kilbertus1 | [
"causality",
"dynamical systems",
"stochastic processes",
"causal discovery",
"signature kernel"
] | causal reasoning | 2402.18477 | 30 | accept (spotlight) | accept | [
8,
6,
10,
8
] | 4 | <thread>
creation date:
1730510163299
thread_type:
official_review
<thread_object>
creator: reviewer_4hzy
<content>
summary:
This paper introduces a novel approach for causal discovery in stochastic processes modeled by stochastic differential equations (SDEs). The authors propose a conditional independence (CI) tes... | metareview:
This paper tackles the challenging and important problem in causal discovery of extracting causal relationships from temporal data governed by a system of stochastic differential equations (SDEs). The authors show how to extract a causal graph from a prototypical system, and propose methods for learning thi... |
z8sxoCYgmd | 2,025 | LOKI: A Comprehensive Synthetic Data Detection Benchmark using Large Multimodal Models | With the rapid development of AI-generated content, the future internet may be inundated with synthetic data, making the discrimination of authentic and credible multimodal data increasingly challenging. Synthetic data detection has thus garnered widespread attention, and the performance of large multimodal models (LMM... | Junyan Ye, Baichuan Zhou, Zilong Huang, Junan Zhang, Tianyi Bai, Hengrui Kang, Jun He, Honglin Lin, Zihao Wang, Tong Wu, Zhizheng Wu, Yiping Chen, Dahua Lin, Conghui He, Weijia Li | ~Junyan_Ye1, ~Baichuan_Zhou1, ~Zilong_Huang2, ~Junan_Zhang1, ~Tianyi_Bai1, ~Hengrui_Kang1, ~Jun_He7, ~Honglin_Lin2, ~Zihao_Wang37, ~Tong_Wu2, ~Zhizheng_Wu1, ~Yiping_Chen2, ~Dahua_Lin1, ~Conghui_He2, ~Weijia_Li2 | [
"lmms；deepfake；multimodality"
] | datasets and benchmarks | 2410.09732 | 49 | accept (spotlight) | accept | [
8,
8,
8,
8
] | 4 | <thread>
creation date:
1729740323010
thread_type:
official_review
<thread_object>
creator: reviewer_qo6r
<content>
summary:
The paper introduces LOKI, a comprehensive benchmark designed to evaluate the capabilities of Large Multimodal Models in detecting synthetic data across multiple modalities. Recognizing the ra... | metareview:
4x accept. This paper introduces a multimodal benchmark aimed at evaluating LMMs’ abilities to detect synthetic data across video, image, 3D, text, and audio modalities. The reviewers agree on the (1) comprehensive coverage of multiple data domains and tasks, (2) clear and well-structured presentation, (3) ... |
Kpjvm2mB0K | 2,025 | Streaming Algorithms For $\ell_p$ Flows and $\ell_p$ Regression | We initiate the study of one-pass streaming algorithms for underdetermined $\ell_p$ linear regression problems of the form
$$
\min_{\mathbf A\mathbf x = \mathbf b} \lVert\mathbf x\rVert_p \,, \qquad
\text{where } \mathbf A \in \mathbb R^{n \times d} \text{ with } n \ll d \,,
$$
which generalizes basi... | Amit Chakrabarti, Jeffrey Jiang, David Woodruff, Taisuke Yasuda | ~Amit_Chakrabarti2, ~Jeffrey_Jiang2, ~David_Woodruff1, ~Taisuke_Yasuda1 | [
"regression",
"streaming",
"online algorithms",
"flows"
] | optimization | null | accept (spotlight) | accept | [
8,
8,
8
] | 3 | <thread>
creation date:
1729422640207
thread_type:
official_review
<thread_object>
creator: reviewer_fogf
<content>
summary:
The $\ell_p$ regression $\min_{\mathbf{Ax} = \mathbf{b}} ||\mathbf{x}||_p$ is a fundamental problem in machine learning, data science, and numerical linear algebra. When $p = 2$, it is the cla... | metareview:
This paper provides streaming algorithms for solving linear regression problems with an $\ell_p$ constraint. In the streaming model data arrives one coordinate at a time. The reviewers deem the results to be good, and this is definitely publishable.
In my own reading, this seems to be an algorithms focused... | |
et5l9qPUhm | 2,025 | Strong Model Collapse | Within the scaling laws paradigm, which underpins the training of large neural networks like ChatGPT and Llama, we consider a supervised regression setting and establish a strong form of the model collapse phenomenon, a critical performance degradation due to synthetic data in the training corpus. Our results show that... | Elvis Dohmatob, Yunzhen Feng, Arjun Subramonian, Julia Kempe | ~Elvis_Dohmatob1, ~Yunzhen_Feng1, ~Arjun_Subramonian1, ~Julia_Kempe1 | [
"model collapse",
"regression",
"high dimensional asymptotics",
"synthetic data",
"scaling laws"
] | learning theory | 2410.04840 | 77 | accept (spotlight) | accept | [
8,
8,
8
] | 3 | <thread>
creation date:
1729697970602
thread_type:
official_review
<thread_object>
creator: reviewer_g7ed
<content>
summary:
This paper studies the model collapse phenomenon in a supervised regression setting. The authors study whether model collapse is inevitable or whether proper mixing strategies can mitigate it.... | metareview:
This paper received unanimously positive reviews from double blind peer review. During the discussion period, a public comment was posted about the lack of citation and discussion for a related paper, first published at COLM 2024 in October 2024. While ICLR's policy exempts papers published after July 1, 20... |
nGiGXLnKhl | 2,025 | Vision-RWKV: Efficient and Scalable Visual Perception with RWKV-Like Architectures | Transformers have revolutionized computer vision and natural language processing, but their high computational complexity limits their application in high-resolution image processing and long-context analysis. This paper introduces Vision-RWKV (VRWKV), a model that builds upon the RWKV architecture from the NLP field w... | Yuchen Duan, Weiyun Wang, Zhe Chen, Xizhou Zhu, Lewei Lu, Tong Lu, Yu Qiao, Hongsheng Li, Jifeng Dai, Wenhai Wang | ~Yuchen_Duan1, ~Weiyun_Wang2, ~Zhe_Chen10, ~Xizhou_Zhu1, ~Lewei_Lu1, ~Tong_Lu1, ~Yu_Qiao1, ~Hongsheng_Li3, ~Jifeng_Dai1, ~Wenhai_Wang2 | [
"rwkv",
"visual perception",
"linear attention"
] | applications to computer vision, audio, language, and other modalities | 2403.02308 | 154 | accept (spotlight) | accept | [
8,
8,
8
] | 3 | <thread>
creation date:
1730246056295
thread_type:
official_review
<thread_object>
creator: reviewer_weiy
<content>
summary:
The authors describe a Vision-RWKV architecture that employs novel techniques like bi-directional RWKV, quad token shifting method Q-shift etc. These techniques helps Vision-RWKV architecture... | metareview:
(a) Scientific Claims and Findings
The paper presents VISION-RWKV, an adaptation of the RWKV network from NLP to vision tasks, incorporating an RNN-based linear attention mechanism. Reviewer QwNu notes that the authors have made necessary modifications to suit visual tasks, while cUXV highlights the introd... |
TJo6aQb7mK | 2,025 | Surprising Effectiveness of pretraining Ternary Language Model at Scale | Rapid advancements in GPU computational power has outpaced memory capacity and bandwidth growth, creating bottlenecks in Large Language Model (LLM) inference. Post-training quantization is the leading method for addressing memory-related bottlenecks in LLM inference, but it suffers from significant performance degradat... | Ayush Kaushal, Tejas Vaidhya, Arnab Kumar Mondal, Tejas Pandey, Aaryan Bhagat, Irina Rish | ~Ayush_Kaushal1, ~Tejas_Vaidhya1, ~Arnab_Kumar_Mondal1, ~Tejas_Pandey2, ~Aaryan_Bhagat1, ~Irina_Rish1 | [
"large language models",
"low-bit language models",
"quantization-aware training",
"pretraining of large language models",
"and scaling laws"
] | foundation or frontier models, including LLMs | 5 | accept (spotlight) | accept | [
8,
10,
10,
5,
5
] | 5 | <thread>
creation date:
1730279032136
thread_type:
official_review
<thread_object>
creator: reviewer_dqjn
<content>
summary:
This paper investigates the scaling law of low bit-width models, specifically ternary language models (TriLMs). The authors present Spectra LLM, an open suite for the quantization-aware traini... | metareview:
TriLMs offers a more efficient alternative to traditional floating-point models and post-training quantized models in terms of scaling behavior and performance at larger model sizes. The paper presents a thorough evaluation of TriLMs across various aspects, including model size scaling, training dynamics, c... | |
eW4yh6HKz4 | 2,025 | CBQ: Cross-Block Quantization for Large Language Models | Post-training quantization (PTQ) has played a pivotal role in compressing large language models (LLMs) at ultra-low costs. Although current PTQ methods have achieved promising results by addressing outliers and employing layer- or block-wise loss optimization techniques, they still suffer from significant performance d... | Xin Ding, Xiaoyu Liu, Zhijun Tu, Yun Zhang, Wei Li, Jie Hu, Hanting Chen, Yehui Tang, Zhiwei Xiong, Baoqun Yin, Yunhe Wang | ~Xin_Ding4, ~Xiaoyu_Liu6, ~Zhijun_Tu1, ~Yun_Zhang5, ~Wei_Li60, ~Jie_Hu8, ~Hanting_Chen1, ~Yehui_Tang1, ~Zhiwei_Xiong1, ~Baoqun_Yin1, ~Yunhe_Wang1 | [
"large language model compression",
"ultra-low bits precision"
] | other topics in machine learning (i.e., none of the above) | 2312.07950 | 38 | accept (spotlight) | accept | [
8,
10,
8,
6,
6
] | 5 | <thread>
creation date:
1730203933645
thread_type:
official_review
<thread_object>
creator: reviewer_mzis
<content>
summary:
This work introduces Cross-Block Quantization (CBQ), a novel post-training quantization (PTQ) framework targeting large language models (LLMs). CBQ's core innovation is its cross-block depende... | metareview:
The reviewers listed some weaknesses: The evaluation is limited to older models like LLAMA1, LLAMA2, and OPT, with no comparison to newer models like LLAMA3, Mistral, or Falcon. Despite claims that LoRA-Rounding reduces computation, no experiments are provided to support this. The comparison with other meth... |
End of preview. Expand in Data Studio
ICLR 2025 — Submissions, Reviews & Metareviews
ICLR 2025 submissions with metadata, the program decision, reviewer ratings, the human peer-review/rebuttal discussion, and the human metareview. Sourced from OpenReview. Two configs:
iclr2025_2k— the curated 1,962-paper evaluation subset used by ReviewerToo (the exact set behind Meta-all 83.4%).iclr2025_full— the full 11,672-paper ICLR 2025 submission pool.
Decision breakdown
iclr2025_2k |
iclr2025_full |
|
|---|---|---|
| Accept (Oral) | 213 | 213 |
| Accept (Spotlight) | 379 | 380 |
| Accept (Poster) | 300 | 3,115 |
| Reject | 500 | 7,893 |
| Withdrawn | 500 | — |
| Desk rejected | 70 | 71 |
| accept / reject | 892 / 1,070 | 3,708 / 7,964 |
Reviews are present for 1,812 / 11,520 papers respectively; metareviews for 1,423 / 8,727 (rejected and withdrawn papers often have neither).
Fields
Both configs share one schema, one row per paper:
| field | type | description |
|---|---|---|
paper_id |
string | OpenReview forum id |
year |
int | 2025 |
title, abstract |
string | submission title and abstract |
authors, author_ids |
string | author names and OpenReview ids |
keywords |
list[string] | author keywords |
primary_area |
string | ICLR primary area |
arxiv_id |
string | arXiv id when matched (else empty) |
citations_serper |
int | citation count at collection time (nullable) |
decision |
string | program decision (raw, lowercased) |
decision_binary |
string | accept / reject |
reviewer_scores |
list[int] | per-reviewer overall ratings |
num_reviewers |
int | number of review threads |
reviews |
string | human peer reviews and author rebuttals, flattened |
metareview |
string | human area-chair metareview |
reviews holds the official reviews and the author/reviewer discussion;
metareview holds the area chair's metareview on its own.
Collection
scrape_iclr2025.py (included) pulls the submissions, decisions, and discussion
from the OpenReview API. Supply your own credentials via the
OPENREVIEW_USERNAME and OPENREVIEW_PASSWORD environment variables.
Loading
from datasets import load_dataset
twok = load_dataset("demfier/reviewertoo-iclr2025-2k", "iclr2025_2k", split="train")
full = load_dataset("demfier/reviewertoo-iclr2025-2k", "iclr2025_full", split="train")
License
Content originates from OpenReview (ICLR 2025), released under CC BY 4.0.
- Downloads last month
- 258