vimarsh244/hf-ICML-2026-agent-repro

Dumping files and logs from : https://huggingface.co/ICML-2026-agent-repro

stars

commits

HTML

primary language

Jul 29, 2026

updated

README

Let's reproduce ICML 2026, together.

This repo is an experiment I am doing of how much research / implementation can be done with AI agents.

This is part of: https://huggingface.co/spaces/ICML-2026-agent-repro/challenge

My thoughts

A good motivation to reproduce a lot of research papers which get published and also validate them. Unfortunately, this is heavily skewed towards those who have lot of capital in terms of tokens they can use and compute for such reproductions.

But such community efforts are super helpful for future researchers and community in large to know which papers have validated claims. Will bring ML conferences closer to those like Usenix with these reproductions.

Also, the models and harnesses have become really excellent. Being able to reproduce papers in parallel and accurately with an judge is excellent and really benefits research - fastracks one to do quick research as tasks like validating previous work and rewriting the code can now done very easily with agents.

Reproductions

34 papers reproduced so far. Each one has a live Trackio logbook on the Hub walking through the claims, the runs behind them, and where the reproduction agreed or diverged from the paper. Code links to the reproduction folder in this repo.

Paper	Logbook	Code
Stochastic Sparse Attention for Memory-Bound Inference _{arXiv:2605.01910}	open	`repro_28058_stochastic_sparse_attention`
Inductive Bias and Spectral Properties of Single-Head Attention in High Dimensions _{arXiv:2509.24914}	open	`repro_17811_single_head_attention_highdim`
Federated Bilevel Performative Prediction	open	`2176_Federated_Bilevel_Performative_Prediction`
Adaptive Personalized Federated Learning via Multi-task Averaging of Kernel Mean Embeddings _{arXiv:2603.02233}	open	`repro_17853_adaptive_personalized_federated_learning`
Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection _{arXiv:2602.03216}	open	`repro_31006_token_sparse_attention`
Connecting Independently Trained Modes via Layer-Wise Connectivity _{arXiv:2505.02604}	open	`repro_16849`
Turning Stale Gradients into Stable Gradients: Coherent Coordinate Descent with Implicit Landscape Smoothing for Lightweight Zeroth-Order Optimization _{arXiv:2605.14373}	open	`repro_26555_cocd_zeroth_order`
Any3D-VLA: Enhancing VLA Robustness via Diverse Point Clouds _{arXiv:2602.00807}	open	`643`
Leveraging Gauge Freedom for Learning Non-Gradient Population Dynamics of Stochastic Systems _{arXiv:2605.25107}	open	`28679`
Full-Batch Gradient Descent Outperforms One-Pass SGD: Sample Complexity Separation in Single-Index Learning _{arXiv:2602.02431}	open	`repro_26332_full-batch-GD-outperforms-one-pass-sgd`
RMNP: Row-Momentum Normalized Preconditioning for Scalable Matrix-Based Optimization _{arXiv:2603.20527}	open	`repro_21319_rmnp_row_momentum_normalized_preconditioning`
REViT: Roto-reflection Equivariant Convolutional Vision Transformer	open	`repro_10620_revit_roto_reflection_equivariant_vit`
Little By Little: Continual Learning via Incremental Mixture of Rank-1 Associative Memory Experts _{arXiv:2506.21035}	open	`repro_10925_moram_continual_rank1_associative_memory`
ToaSt: Token Channel Selection and Structured Pruning for Efficient ViT _{arXiv:2602.15720}	open	`repro_1840_toast_token_channel_pruning_vit`
FedFit: Federated Dynamic Sparse Training via Fisher Information scoring	open	`repro_15844_fedfit_federated_fisher_sparse_training`
Event2Vec — Processing Neuromorphic Events Directly by Representations in Vector Space _{arXiv:2504.15371}	open	`repro_2856_event2vec_neuromorphic_vector_space`
A Capacity-Based Rationale for Multi-Head Attention _{arXiv:2509.22840}	open	`repro_26081_capacity_rationale_multihead_attention`
Sample Margin-Aware Recalibration of Temperature Scaling _{arXiv:2506.23492}	open	`repro_5500_smart_margin_aware_temperature_scaling`
How Can Mamba Learn In Context with Outliers and Generalize Provably? _{arXiv:2510.00399}	open	`repro_20925_mamba_icl_outliers`
Deep Networks Learn Deep Hierarchical Models _{arXiv:2601.00455}	open	`repro_12249_deep_hierarchical_models`
Understanding Dynamics of Adam in Zero-Sum Games: An ODE Approach _{arXiv:2605.19392}	open	`repro_understanding_dynamics_adam_zero_sum_games`
Rational Transductors _{arXiv:2602.07599}	open	`repro_rational_transductors`
Quantifying the noise sensitivity of the Wasserstein metric for images _{arXiv:2510.01015}	open	`repro_noise_sensitivity_wasserstein`
On the Theory of Continual Learning with Gradient Descent for Neural Networks _{arXiv:2510.05573}	open	`repro_17723`
How Many Different Outputs Can a Transformer Generate?	open	`repro_how_many_different_outputs_can_a_transformer_generate`
Delving into Muon and Beyond: Deep Analysis and Extensions _{arXiv:2602.04669}	open	`repro_14014_muon_beyond`
Combinatorial Sparse PCA Beyond the Spiked Identity Model _{arXiv:2603.02607}	open	`repro_combinatorial_sparse_pca_beyond_the_spiked_identity_model`
Adversarially Robust Approximate Furthest Neighbor _{arXiv:2605.16618}	open	`repro_28185_robust_afn`
Why Are Linear RNNs More Parallelizable?	open	`repro_18268`
SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity _{arXiv:2603.05232}	open	`slidesparse_icml33765`
Float8@2bits: Entropy Coding Enables Data-Free Model Compression _{arXiv:2601.22787}	open	`repro_float8_2bits`
Efficient Diffusion Models under Nonconvex Equality and Inequality constraints via Landing _{arXiv:2604.17838}	open	`repro_32912`
Cost-aware Stopping for Bayesian Optimization _{arXiv:2507.12453}	open	`repro_6861`
MIMOMamba: From Scalar Duality to Matrix-Valued Attention	open	`repro_11253_mimomamba`

All logbooks live under huggingface.co/vimarsh and are tagged icml2026-repro.

Contributors

vimarsh244

18 commits

Similar repos

vimarsh/repro-stochastic-sparse-attention-for-memory-bound-inference

Space

morganstanley/MSML

Repo for Morgan Stanley Machine Learning Research group's publications

177

vimarsh/repro-fedfit-federated-dynamic-sparse-training-via-fisher-information-scoring

Space

vimarsh/repro-a-capacity-based-rationale-for-multi-head-attention

Space

vimarsh/repro-adaptive-personalized-fl-multi-task-averaging-kernel-mean-embedding

Space

vimarsh/repro-token-sparse-attention-long-context-interleaved-selection

Space

tsinghua-fib-lab/AutoSOTA

674

vimarsh/repro-toast-token-channel-selection-and-structured-pruning-for-efficient-vit

Space

Languages

HTML

90.6%

Jupyter Notebook

4.0%

Python

3.8%

vimarsh244/hf-ICML-2026-agent-repro

Dumping files and logs from : https://huggingface.co/ICML-2026-agent-repro

stars

commits

HTML

primary language

Jul 29, 2026

updated

README

Let's reproduce ICML 2026, together.

This repo is an experiment I am doing of how much research / implementation can be done with AI agents.

This is part of: https://huggingface.co/spaces/ICML-2026-agent-repro/challenge

My thoughts

Reproductions

Paper	Logbook	Code
Stochastic Sparse Attention for Memory-Bound Inference _{arXiv:2605.01910}	open	`repro_28058_stochastic_sparse_attention`
Inductive Bias and Spectral Properties of Single-Head Attention in High Dimensions _{arXiv:2509.24914}	open	`repro_17811_single_head_attention_highdim`
Federated Bilevel Performative Prediction	open	`2176_Federated_Bilevel_Performative_Prediction`
Adaptive Personalized Federated Learning via Multi-task Averaging of Kernel Mean Embeddings _{arXiv:2603.02233}	open	`repro_17853_adaptive_personalized_federated_learning`
Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection _{arXiv:2602.03216}	open	`repro_31006_token_sparse_attention`
Connecting Independently Trained Modes via Layer-Wise Connectivity _{arXiv:2505.02604}	open	`repro_16849`
Turning Stale Gradients into Stable Gradients: Coherent Coordinate Descent with Implicit Landscape Smoothing for Lightweight Zeroth-Order Optimization _{arXiv:2605.14373}	open	`repro_26555_cocd_zeroth_order`
Any3D-VLA: Enhancing VLA Robustness via Diverse Point Clouds _{arXiv:2602.00807}	open	`643`
Leveraging Gauge Freedom for Learning Non-Gradient Population Dynamics of Stochastic Systems _{arXiv:2605.25107}	open	`28679`
Full-Batch Gradient Descent Outperforms One-Pass SGD: Sample Complexity Separation in Single-Index Learning _{arXiv:2602.02431}	open	`repro_26332_full-batch-GD-outperforms-one-pass-sgd`
RMNP: Row-Momentum Normalized Preconditioning for Scalable Matrix-Based Optimization _{arXiv:2603.20527}	open	`repro_21319_rmnp_row_momentum_normalized_preconditioning`
REViT: Roto-reflection Equivariant Convolutional Vision Transformer	open	`repro_10620_revit_roto_reflection_equivariant_vit`
Little By Little: Continual Learning via Incremental Mixture of Rank-1 Associative Memory Experts _{arXiv:2506.21035}	open	`repro_10925_moram_continual_rank1_associative_memory`
ToaSt: Token Channel Selection and Structured Pruning for Efficient ViT _{arXiv:2602.15720}	open	`repro_1840_toast_token_channel_pruning_vit`
FedFit: Federated Dynamic Sparse Training via Fisher Information scoring	open	`repro_15844_fedfit_federated_fisher_sparse_training`
Event2Vec — Processing Neuromorphic Events Directly by Representations in Vector Space _{arXiv:2504.15371}	open	`repro_2856_event2vec_neuromorphic_vector_space`
A Capacity-Based Rationale for Multi-Head Attention _{arXiv:2509.22840}	open	`repro_26081_capacity_rationale_multihead_attention`
Sample Margin-Aware Recalibration of Temperature Scaling _{arXiv:2506.23492}	open	`repro_5500_smart_margin_aware_temperature_scaling`
How Can Mamba Learn In Context with Outliers and Generalize Provably? _{arXiv:2510.00399}	open	`repro_20925_mamba_icl_outliers`
Deep Networks Learn Deep Hierarchical Models _{arXiv:2601.00455}	open	`repro_12249_deep_hierarchical_models`
Understanding Dynamics of Adam in Zero-Sum Games: An ODE Approach _{arXiv:2605.19392}	open	`repro_understanding_dynamics_adam_zero_sum_games`
Rational Transductors _{arXiv:2602.07599}	open	`repro_rational_transductors`
Quantifying the noise sensitivity of the Wasserstein metric for images _{arXiv:2510.01015}	open	`repro_noise_sensitivity_wasserstein`
On the Theory of Continual Learning with Gradient Descent for Neural Networks _{arXiv:2510.05573}	open	`repro_17723`
How Many Different Outputs Can a Transformer Generate?	open	`repro_how_many_different_outputs_can_a_transformer_generate`
Delving into Muon and Beyond: Deep Analysis and Extensions _{arXiv:2602.04669}	open	`repro_14014_muon_beyond`
Combinatorial Sparse PCA Beyond the Spiked Identity Model _{arXiv:2603.02607}	open	`repro_combinatorial_sparse_pca_beyond_the_spiked_identity_model`
Adversarially Robust Approximate Furthest Neighbor _{arXiv:2605.16618}	open	`repro_28185_robust_afn`
Why Are Linear RNNs More Parallelizable?	open	`repro_18268`
SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity _{arXiv:2603.05232}	open	`slidesparse_icml33765`
Float8@2bits: Entropy Coding Enables Data-Free Model Compression _{arXiv:2601.22787}	open	`repro_float8_2bits`
Efficient Diffusion Models under Nonconvex Equality and Inequality constraints via Landing _{arXiv:2604.17838}	open	`repro_32912`
Cost-aware Stopping for Bayesian Optimization _{arXiv:2507.12453}	open	`repro_6861`
MIMOMamba: From Scalar Duality to Matrix-Valued Attention	open	`repro_11253_mimomamba`