paper_id stringlengths 5 5 | paper_title stringlengths 8 128 | stem stringlengths 20 126 | pdf unknown | methodology_text stringlengths 0 5k | num_images int32 0 178 | images images listlengths 0 178 | captions listlengths 0 178 | bboxes listlengths 0 178 | page_indices listlengths 0 178 | content_list large_stringlengths 67k 1.35M | markdown large_stringlengths 46.1k 861k |
|---|---|---|---|---|---|---|---|---|---|---|---|
00013 | Differentiable Sparsity via $D$ -Gating: Simple and Versatile Structured Penalization | 00013_Differentiable_Sparsity_via_D-Gating_Simple_and_Versatile_Structured_Penalization | "JVBERi0xLjUKJb/3ov4KMTk2MyAwIG9iago8PCAvTGluZWFyaXplZCAxIC9MIDE0MTg5MjEgL0ggWyAxMjIxMCAxMjE5IF0gL08(...TRUNCATED) | "Inspired by prior work on differentiable sparse regularization, we propose a new approach called $D(...TRUNCATED) | 22 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Parameter trajectories for a two-feature squared loss toy objective with non-convex $\\b(...TRUNCATED) | ["[258, 90, 741, 222]","[178, 92, 821, 215]","[178, 90, 823, 193]","[179, 94, 818, 198]","[176, 661,(...TRUNCATED) | [
1,
2,
5,
6,
6,
7,
8,
8,
24,
24,
25,
25,
26,
26,
26,
27,
27,
27,
28,
29,
29,
30
] | "[\n {\n \"type\": \"text\",\n \"text\": \"Differentiable Sparsity via $D$ -Gating: Simple an(...TRUNCATED) | "# Differentiable Sparsity via $D$ -Gating: Simple and Versatile Structured Penalization\n\nChris $\(...TRUNCATED) |
00028 | ReSim: Reliable World Simulation for Autonomous Driving | 00028_ReSim_Reliable_World_Simulation_for_Autonomous_Driving | "JVBERi0xLjUKJb/3ov4KMTUyNyAwIG9iago8PCAvTGluZWFyaXplZCAxIC9MIDg5NDg1MTcgL0ggWyAyOTQxIDU3OSBdIC9PIDE(...TRUNCATED) | "Basics. ReSim is built on CogVideoX [30], a high-capacity diffusion transformer originally conditio(...TRUNCATED) | 18 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Overview of ReSim. (a) Heterogeneous driving data includes (i,ii) experts’ safe drivin(...TRUNCATED) | ["[179, 87, 821, 404]","[478, 239, 816, 376]","[173, 707, 818, 869]","[181, 94, 464, 200]","[174, 89(...TRUNCATED) | [
1,
4,
4,
5,
6,
6,
7,
8,
8,
8,
25,
27,
29,
29,
30,
30,
31,
31
] | "[\n {\n \"type\": \"text\",\n \"text\": \"ReSim: Reliable World Simulation for Autonomous Dr(...TRUNCATED) | "# ReSim: Reliable World Simulation for Autonomous Driving\n\nJiazhi Yang1,3 Kashyap Chitta4,7 Sheny(...TRUNCATED) |
00066 | Do-PFN: In-Context Learning for Causal Effect Estimation | 00066_Do-PFN_In-Context_Learning_for_Causal_Effect_Estimation | "JVBERi0xLjUKJb/3ov4KOTIyIDAgb2JqCjw8IC9MaW5lYXJpemVkIDEgL0wgNTQ5NTg2MCAvSCBbIDI3NTUgNzg2IF0gL08gOTI(...TRUNCATED) | "1. Do-PFN: We propose Do-PFN, a foundation model pre-trained on data from structural causal models (...TRUNCATED) | 22 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Do-PFN overview: Do-PFN performs in-context learning (ICL) for causal effect estimation,(...TRUNCATED) | ["[194, 92, 800, 300]","[173, 87, 825, 178]","[174, 85, 825, 338]","[174, 89, 823, 219]","[174, 753,(...TRUNCATED) | [
1,
6,
7,
8,
8,
9,
23,
23,
24,
25,
25,
26,
27,
27,
27,
28,
28,
29,
30,
30,
30,
30
] | "[\n {\n \"type\": \"text\",\n \"text\": \"Do-PFN: In-Context Learning for Causal Effect Esti(...TRUNCATED) | "# Do-PFN: In-Context Learning for Causal Effect Estimation\n\nJake Robertson1,2,3∗ Arik Reuter4,5(...TRUNCATED) |
00093 | "$\\Psi$ -Sampler: Initial Particle Sampling for SMC-Based Inference-Time Reward Alignment in Score (...TRUNCATED) | "00093_Psi-Sampler_Initial_Particle_Sampling_for_SMC-Based_Inference-Time_Reward_Alignment_in_Score_(...TRUNCATED) | "JVBERi0xLjUKJb/3ov4KMTE0NiAwIG9iago8PCAvTGluZWFyaXplZCAxIC9MIDQxMDM3MDkwIC9IIFsgMjg0MyA4MDYgXSAvTyA(...TRUNCATED) | "Taehoon Yoon∗ Yunhong Min∗ Kyeongmin Yeo∗ Minhyuk Sung KAIST {taehoon,dbsghd363,aaaaa,mhsung}(...TRUNCATED) | 16 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Toy sampling–method comparison. Each panel visualizes both the initial samples (blue) (...TRUNCATED) | ["[176, 85, 823, 196]","[174, 154, 831, 827]","[176, 88, 823, 181]","[174, 569, 823, 844]","[178, 11(...TRUNCATED) | [
6,
8,
9,
28,
31,
33,
33,
33,
33,
33,
33,
33,
33,
34,
35,
36
] | "[\n {\n \"type\": \"text\",\n \"text\": \"$\\\\Psi$ -Sampler: Initial Particle Sampling for (...TRUNCATED) | "# $\\Psi$ -Sampler: Initial Particle Sampling for SMC-Based Inference-Time Reward Alignment in Scor(...TRUNCATED) |
00106 | What Makes a Reward Model a Good Teacher? An Optimization Perspective | 00106_What_Makes_a_Reward_Model_a_Good_Teacher_An_Optimization_Perspective | "JVBERi0xLjUKJdDUxdgKMTk0IDAgb2JqCjw8Ci9MZW5ndGggMzY0MiAgICAgIAovRmlsdGVyIC9GbGF0ZURlY29kZQo+PgpzdHJ(...TRUNCATED) | "Noam Razin, Zixuan Wang, Hubert Strauss, Stanley Wei, Jason D. Lee, Sanjeev Arora \nPrinceton Langu(...TRUNCATED) | 14 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Illustration of how accuracy (Definition 1) and reward variance (Definition 2) affect th(...TRUNCATED) | ["[178, 80, 821, 297]","[176, 282, 816, 431]","[176, 308, 818, 444]","[176, 94, 816, 244]","[176, 50(...TRUNCATED) | [
1,
7,
8,
53,
53,
53,
54,
54,
55,
56,
56,
57,
57,
57
] | "[\n {\n \"type\": \"text\",\n \"text\": \"What Makes a Reward Model a Good Teacher? An Optim(...TRUNCATED) | "# What Makes a Reward Model a Good Teacher? An Optimization Perspective\n\nNoam Razin, Zixuan Wang,(...TRUNCATED) |
00189 | Q-Insight: Understanding Image Quality via Visual Reinforcement Learning | 00189_Q-Insight_Understanding_Image_Quality_via_Visual_Reinforcement_Learning | "JVBERi0xLjUKJb/3ov4KNjUxIDAgb2JqCjw8IC9MaW5lYXJpemVkIDEgL0wgMjY0NDUzNyAvSCBbIDI3MDEgNDY2IF0gL08gNjU(...TRUNCATED) | "Group Relative Policy Optimization (GRPO) is an innovative reinforcement learning paradigm that has(...TRUNCATED) | 23 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: PLCC comparisons between our proposed Q-Insight and existing IQA metrics (left) and thre(...TRUNCATED) | ["[173, 89, 820, 222]","[178, 87, 821, 275]","[173, 88, 377, 324]","[759, 215, 821, 266]","[191, 349(...TRUNCATED) | [
1,
3,
6,
6,
7,
8,
8,
17,
17,
17,
17,
17,
17,
17,
17,
17,
18,
18,
18,
18,
18,
18,
18
] | "[\n {\n \"type\": \"text\",\n \"text\": \"Q-Insight: Understanding Image Quality via Visual (...TRUNCATED) | "# Q-Insight: Understanding Image Quality via Visual Reinforcement Learning\n\nWeiqi ${ \\bf { U } }(...TRUNCATED) |
00232 | GraphMaster: Automated Graph Synthesis via LLM Agents in Data-Limited Environments | 00232_GraphMaster_Automated_Graph_Synthesis_via_LLM_Agents_in_Data-Limited_Environments | "JVBERi0xLjUKJb/3ov4KMjU0NSAwIG9iago8PCAvTGluZWFyaXplZCAxIC9MIDMwNTgyNjQgL0ggWyAzNTUxIDExMjMgXSAvTyA(...TRUNCATED) | "Traditional graph data synthesis methods [7] address data scarcity through various approaches. Edge(...TRUNCATED) | 10 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: GraphMaster: A hierarchical multi-agent framework for text-attributed graph synthesis. "(...TRUNCATED) | ["[174, 88, 825, 265]","[184, 236, 815, 367]","[196, 88, 802, 224]","[181, 409, 816, 669]","[183, 85(...TRUNCATED) | [
3,
7,
8,
28,
29,
29,
30,
30,
31,
43
] | "[\n {\n \"type\": \"text\",\n \"text\": \"GraphMaster: Automated Graph Synthesis via LLM Age(...TRUNCATED) | "# GraphMaster: Automated Graph Synthesis via LLM Agents in Data-Limited Environments\n\nEnjun $\\ma(...TRUNCATED) |
00251 | BIOCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning | 00251_BioCLIP_2_Emergent_Properties_from_Scaling_Hierarchical_Contrastive_Learning | "JVBERi0xLjUKJb/3ov4KMTI1OSAwIG9iago8PCAvTGluZWFyaXplZCAxIC9MIDEzMDgwMzQxIC9IIFsgMzA5NSA4ODEgXSAvTyA(...TRUNCATED) | 15 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: While BIOCLIP 2 is trained to distinguish species, it demonstrates emergent properties b(...TRUNCATED) | ["[174, 348, 820, 493]","[174, 90, 816, 224]","[176, 299, 816, 468]","[181, 97, 820, 218]","[181, 97(...TRUNCATED) | [
0,
2,
5,
6,
7,
8,
24,
25,
25,
26,
29,
31,
31,
32,
32
] | "[\n {\n \"type\": \"text\",\n \"text\": \"BIOCLIP 2: Emergent Properties from Scaling Hierar(...TRUNCATED) | "# BIOCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning\n\nJianyang $\\math(...TRUNCATED) | |
00274 | Approximate Domain Unlearning for Vision-Language Models | 00274_Approximate_Domain_Unlearning_for_Vision-Language_Models | "JVBERi0xLjUKJb/3ov4KNzI3IDAgb2JqCjw8IC9MaW5lYXJpemVkIDEgL0wgMTg2OTcwOCAvSCBbIDI4ODUgNjAzIF0gL08gNzM(...TRUNCATED) | "Kodai Kawamura∗1,2, Yuta Goto∗1, Rintaro Yanagi3, Hirokatsu Kataoka3,4, Go Irie1 \n1Tokyo Unive(...TRUNCATED) | 9 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Illustration of Approximate Domain Unlearning (ADU). ADU is a novel approximate unlearni(...TRUNCATED) | ["[504, 175, 813, 334]","[176, 92, 823, 271]","[222, 90, 776, 219]","[178, 305, 820, 422]","[230, 11(...TRUNCATED) | [
1,
2,
7,
7,
8,
9,
26,
27,
28
] | "[\n {\n \"type\": \"text\",\n \"text\": \"Approximate Domain Unlearning for Vision-Language (...TRUNCATED) | "# Approximate Domain Unlearning for Vision-Language Models\n\nKodai Kawamura∗1,2, Yuta Goto∗1, (...TRUNCATED) |
00279 | VoxDet: Rethinking 3D Semantic Scene Completion as Dense Object Detection | 00279_VoxDet_Rethinking_3D_Semantic_Scene_Completion_as_Dense_Object_Detection | "JVBERi0xLjUKJb/3ov4KNzExIDAgb2JqCjw8IC9MaW5lYXJpemVkIDEgL0wgMjEyNjM4MjkgL0ggWyAzNDgxIDcwMCBdIC9PIDc(...TRUNCATED) | "Overview. Fig. 3 shows the overall workflow of our VoxDet. Given RGB input, we follow previous work(...TRUNCATED) | 17 | [{"src":"https://huggingface.co/proxy/datasets-server.huggingface.co/assets/Samarth0710/neurips2025-papers/--/{dataset_gi(...TRUNCATED) | ["Figure 1: Schematic comparison of previous SSC paradigm [6, 2, 79] and the proposed VoxDet. Left: (...TRUNCATED) | ["[176, 89, 823, 200]","[174, 90, 563, 229]","[174, 89, 821, 203]","[191, 89, 816, 188]","[176, 500,(...TRUNCATED) | [
1,
3,
4,
5,
7,
8,
9,
23,
25,
26,
26,
26,
26,
28,
30,
33,
34
] | "[\n {\n \"type\": \"text\",\n \"text\": \"VoxDet: Rethinking 3D Semantic Scene Completion as(...TRUNCATED) | "# VoxDet: Rethinking 3D Semantic Scene Completion as Dense Object Detection\n\nWuyang Li1 Zhu $\\ma(...TRUNCATED) |
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 10