ML Research Hub
32.3K subscribers
6.74K photos
472 videos
24 files
7.35K links
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.

Admin: @HusseinSheikho || @Hussein_Sheikho
Download Telegram
Process Reward Agents for Steering Knowledge-Intensive Reasoning

📝 Summary:
Process Reward Agents provide domain-grounded, online step-wise rewards for frozen policies in knowledge-intensive reasoning, enabling improved search-based decoding and generalizing across different ...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09482
• PDF: https://arxiv.org/pdf/2604.09482
• Project Page: https://process-reward-agents.github.io/

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Robust Reasoning Benchmark

📝 Summary:
Research reveals that large language models exhibit fragile reasoning capabilities when subjected to perturbations, with open-weight models showing significant accuracy drops and evidence of memory po...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.08571
• PDF: https://arxiv.org/pdf/2604.08571

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Cactus: Accelerating Auto-Regressive Decoding with Constrained Acceptance Speculative Sampling

📝 Summary:
Speculative sampling methods are enhanced by formulating them as constrained optimization problems, enabling controlled distribution divergence while maintaining high acceptance rates and output quali...

🔹 Publication Date: Published on Apr 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04987
• PDF: https://arxiv.org/pdf/2604.04987
• Github: https://github.com/MANGA-UOFA/Cactus

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
MixFlow: Mixed Source Distributions Improve Rectified Flows

📝 Summary:
Rectified flows and diffusion models are improved through κ-FC formulation that conditions the source distribution and MixFlow training strategy that reduces generative path curvatures and enhances sa...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09181
• PDF: https://arxiv.org/pdf/2604.09181
• Github: https://github.com/NazirNayal8/MixFlow

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#RectifiedFlows #DiffusionModels #GenerativeAI #MachineLearning #AIResearch
Beyond the Assistant Turn: User Turn Generation as a Probe of Interaction Awareness in Language Models

📝 Summary:
User-turn generation probes LLM interaction awareness, decoupled from task accuracy. This awareness is often latent but revealed by higher temperature sampling and can be improved through post-training, uncovering a new dimension of LLM behavior.

🔹 Publication Date: Published on Apr 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02315
• PDF: https://arxiv.org/pdf/2604.02315

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#LLM #NLP #AI #InteractionAwareness #UserTurnGeneration
Pseudo-Unification: Entropy Probing Reveals Divergent Information Patterns in Unified Multimodal Models

📝 Summary:
Unified multimodal models suffer from pseudo-unification due to asymmetric encoding and split response patterns, requiring consistent information flow for genuine multimodal synergy. AI-generated summ...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10949
• PDF: https://arxiv.org/pdf/2604.10949

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
CodeTracer: Towards Traceable Agent States

📝 Summary:
CodeTracer is a tracing architecture that analyzes code agent execution by reconstructing state transitions and localizing failures in complex multi-stage workflows. AI-generated summary Code agents a...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11641
• PDF: https://arxiv.org/pdf/2604.11641

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Media is too big
VIEW IN TELEGRAM
OmniShow: Unifying Multimodal Conditions for Human-Object Interaction Video Generation

📝 Summary:
OmniShow is an end-to-end framework for human-object interaction video generation using multimodal conditions like text, images, audio, and pose. It uses unified conditioning, gated attention, and decoupled training to achieve state-of-the-art performance despite data scarcity. A new benchmark, H...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2604.11804
• PDF: https://arxiv.org/pdf/2604.11804
• Project Page: https://correr-zhou.github.io/OmniShow
• Github: https://github.com/Correr-Zhou/OmniShow

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
TAIHRI: Task-Aware 3D Human Keypoints Localization for Close-Range Human-Robot Interaction

📝 Summary:
TAIHRI is a vision-language model designed for egocentric human-robot interaction that enables precise 3D keypoint localization through 2D keypoint reasoning and next token prediction. AI-generated su...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.08921
• PDF: https://arxiv.org/pdf/2604.08921
• Github: https://github.com/Tencent/TAIHRI

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
General365: Benchmarking General Reasoning in Large Language Models Across Diverse and Challenging Tasks

📝 Summary:
Large language models demonstrate limited general reasoning capabilities despite strong domain-specific performance, as revealed by a new benchmark assessing K-12 level reasoning across diverse proble...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11778
• PDF: https://arxiv.org/pdf/2604.11778
• Project Page: https://general365.github.io/
• Github: https://general365.github.io/

Datasets citing this paper:
https://huggingface.co/datasets/meituan-longcat/General365_Public

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration

📝 Summary:
A nonlinear extrapolation framework for reinforcement learning with verifiable rewards in large language models that reduces computational overhead by modeling rank-1 parameter trajectories through Lo...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11446
• PDF: https://arxiv.org/pdf/2604.11446
• Github: https://github.com/RUCAIBox/NExt

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
From Reasoning to Agentic: Credit Assignment in Reinforcement Learning for Large Language Models

📝 Summary:
Credit assignment methods for reinforcement learning with large language models are categorized by granularity and methodology, with distinct approaches emerging for reasoning versus agentic settings....

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09459
• PDF: https://arxiv.org/pdf/2604.09459
• Github: https://github.com/xxzcc/Awesome-Credit-Assignment-in-LLM-RL

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
ADD for Multi-Bit Image Watermarking

📝 Summary:
ADD is a multi-bit image watermarking method that uses linear combination and inner product operations for embedding and decoding, achieving high accuracy and efficiency compared to existing approache...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11491
• PDF: https://arxiv.org/pdf/2604.11491

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
TorchUMM: A Unified Multimodal Model Codebase for Evaluation, Analysis, and Post-training

📝 Summary:
TorchUMM presents a unified codebase for evaluating and analyzing multimodal models across understanding, generation, and editing tasks with standardized protocols and diverse datasets. AI-generated s...

🔹 Publication Date: Published on Apr 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10784
• PDF: https://arxiv.org/pdf/2604.10784
• Project Page: https://aifrontierlab.github.io/TorchUMM/
• Github: https://github.com/AIFrontierLab/TorchUMM

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Media is too big
VIEW IN TELEGRAM
Strips as Tokens: Artist Mesh Generation with Native UV Segmentation

📝 Summary:
SATO introduces a novel token ordering strategy for autoregressive transformers that preserves edge flow and semantic layout in mesh generation through triangle strip-based sequences. AI-generated sum...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09132
• PDF: https://arxiv.org/pdf/2604.09132
• Project Page: https://ruixu.me/html/SATO/index.html
• Github: https://github.com/Xrvitd/SATO

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation

📝 Summary:
Transformers face challenges from Attention Sink phenomenon where excessive attention focuses on uninformative tokens, impacting interpretability and performance, necessitating comprehensive research ...

🔹 Publication Date: Published on Apr 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10098
• PDF: https://arxiv.org/pdf/2604.10098
• Github: https://github.com/ZunhaiSu/Awesome-Attention-Sink

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
CocoaBench: Evaluating Unified Digital Agents in the Wild

📝 Summary:
A new benchmark called CocoaBench evaluates unified digital agents on complex, multi-capability tasks requiring vision, search, and coding integration, revealing significant room for improvement in cu...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11201
• PDF: https://arxiv.org/pdf/2604.11201
• Project Page: https://cocoabench.github.io/
• Github: https://github.com/cocoabench/cocoa-agent

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents

📝 Summary:
ClawGUI is an open-source framework that unifies reinforcement learning training, standardized evaluation, and cross-platform deployment for GUI agents. It provides infrastructure for virtual and real environments, consistent benchmarks, and agent deployment to mobile devices. ClawGUI improves GU...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11784
• PDF: https://arxiv.org/pdf/2604.11784
• Project Page: https://zju-real.github.io/ClawGUI-Page/

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Introspective Diffusion Language Models

📝 Summary:
Introspective Diffusion Language Models address quality gaps with autoregressive models by enforcing introspective consistency through novel decoding algorithms and optimized inference engines. AI-gen...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11035
• PDF: https://arxiv.org/pdf/2604.11035
• Project Page: https://introspective-diffusion.github.io/
• Github: https://github.com/Introspective-Diffusion/I-DLM

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Audio Flamingo Next: Next-Generation Open Audio-Language Models for Speech, Sound, and Music

📝 Summary:
Audio Flamingo Next represents a significant advancement in audio-language modeling with enhanced understanding capabilities, extended audio input lengths, and novel temporal reasoning mechanisms. AI-...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10905
• PDF: https://arxiv.org/pdf/2604.10905
• Project Page: https://afnext-umd-nvidia.github.io/

🔹 Models citing this paper:
https://huggingface.co/nvidia/audio-flamingo-next-hf
https://huggingface.co/nvidia/audio-flamingo-next-captioner-hf
https://huggingface.co/nvidia/audio-flamingo-next-think-hf

Spaces citing this paper:
https://huggingface.co/spaces/nvidia/audio-flamingo-next
https://huggingface.co/spaces/nvidia/audio-flamingo-next-captioner

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research