✨Can LLMs Learn to Reason Robustly under Noisy Supervision?

📝 Summary:
Reinforcement Learning with Verifiable Rewards faces challenges with noisy labels, but a proposed method called Online Label Refinement addresses this by progressively correcting labels based on polic...

🔹 Publication Date: Published on Apr 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03993
• PDF: https://arxiv.org/pdf/2604.03993
• Github: https://github.com/ShenzhiYang2000/OLR

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨HDP: A Lightweight Cryptographic Protocol for Human Delegation Provenance in Agentic AI Systems

📝 Summary:
Agentic AI systems lack verifiable human authorization for delegated tasks. HDP is a lightweight cryptographic protocol that records and verifies the full human delegation provenance using tokens, allowing offline checks.

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04522
• PDF: https://arxiv.org/pdf/2604.04522

✨ Spaces citing this paper:
• https://huggingface.co/spaces/helixar-ai/hdp-physical-demo

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Self-Execution Simulation Improves Coding Models

📝 Summary:
This work trains code LLMs to simulate program execution step-by-step using fine-tuning and reinforcement learning. This enables self-verification and iterative self-fixing, significantly improving competitive programming performance and outperforming standard reasoning methods.

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03253
• PDF: https://arxiv.org/pdf/2604.03253

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

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

#CodeLLMs #AI #ReinforcementLearning #DeepLearning #CompetitiveProgramming

✨AvatarPointillist: AutoRegressive 4D Gaussian Avatarization

📝 Summary:
AvatarPointillist creates dynamic 4D Gaussian avatars from a single image using an autoregressive Transformer. It builds point clouds with adaptive density and binding info for realistic animation, producing high-quality, controllable results.

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04787
• PDF: https://arxiv.org/pdf/2604.04787
• Project Page: https://kumapowerliu.github.io/AvatarPointillist/
• Github: https://github.com/KumapowerLIU/AvatarPointillist

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

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

#AI #ComputerVision #3DAvatars #GenerativeAI #MachineLearning

✨Your Agent, Their Asset: A Real-World Safety Analysis of OpenClaw

📝 Summary:
A real-world safety analysis of the personal AI agent OpenClaw reveals significant vulnerabilities due to its broad system access. Attacks targeting its Capability, Identity, or Knowledge CIK dimensions drastically increase success rates, and current defenses are insufficient, indicating inherent...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04759
• PDF: https://arxiv.org/pdf/2604.04759
• Project Page: https://ucsc-vlaa.github.io/CIK-Bench/
• Github: https://github.com/UCSC-VLAA/CIK-Bench

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

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

#AISafety #Cybersecurity #AIAgents #Vulnerability #AIsecurity

✨Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing

📝 Summary:
SRPO unifies GRPO and SDPO in reinforcement learning by routing correct samples to GRPO's reward-aligned reinforcement and failed samples to SDPO's targeted logit-level correction. This novel approach achieves superior stability, rapid improvement, and better performance than either baseline.

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02288
• PDF: https://arxiv.org/pdf/2604.02288

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

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

#ReinforcementLearning #PolicyOptimization #SampleRouting #MachineLearning #AIResearch

✨LIBERO-Para: A Diagnostic Benchmark and Metrics for Paraphrase Robustness in VLA Models

📝 Summary:
Vision-Language-Action models show significant performance drops when handling paraphrased instructions due to surface-level matching rather than semantic understanding, highlighting the need for bett...

🔹 Publication Date: Published on Mar 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.28301
• PDF: https://arxiv.org/pdf/2603.28301
• Project Page: https://cau-hai-lab.github.io/LIBERO-Para/
• Github: https://github.com/cau-hai-lab/LIBERO-Para

✨ Datasets citing this paper:
• https://huggingface.co/datasets/HAI-Lab/LIBERO-Para

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learning to Learn-at-Test-Time: Language Agents with Learnable Adaptation Policies

📝 Summary:
Meta-TTL formulates adaptation policy discovery as a bi-level optimization problem to improve language agent performance through learned policies rather than hand-crafted ones. AI-generated summary Te...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00830
• PDF: https://arxiv.org/pdf/2604.00830
• Github: https://github.com/zzzlou/meta-ttl

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SciLT: Long-Tailed Classification in Scientific Image Domains

📝 Summary:
Scientific long-tailed recognition benefits from a proposed framework that leverages multi-level representations through adaptive feature fusion and dual-supervision learning to achieve balanced perfo...

🔹 Publication Date: Published on Apr 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03687
• PDF: https://arxiv.org/pdf/2604.03687

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨PLUME: Latent Reasoning Based Universal Multimodal Embedding

📝 Summary:
PLUME introduces a latent reasoning framework for universal multimodal embedding that replaces explicit chain-of-thought reasoning with continuous latent state rollouts, achieving faster inference whi...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02073
• PDF: https://arxiv.org/pdf/2604.02073
• Project Page: https://haoxiangzhao12138.github.io/PLUME/
• Github: https://github.com/haoxiangzhao12138/PLUME

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

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

#MultimodalAI #LatentReasoning #Embeddings #AIResearch #MachineLearning

✨Adam's Law: Textual Frequency Law on Large Language Models

📝 Summary:
Adam's Law proposes a novel framework to improve LLM performance through textual frequency analysis. It introduces Textual Frequency Law for prompting/fine-tuning, Distillation for estimation, and Curriculum Training. Experiments demonstrate its effectiveness.

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02176
• PDF: https://arxiv.org/pdf/2604.02176
• Github: https://github.com/HongyuanLuke/frequencylaw

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

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

#LLM #TextFrequency #PromptEngineering #NLP #DeepLearning

✨CLEAR: Unlocking Generative Potential for Degraded Image Understanding in Unified Multimodal Models

📝 Summary:
CLEAR improves multimodal models robustness to image degradation. It connects the models generative and reasoning capabilities using supervised fine-tuning, a latent representation bridge, and reinforcement learning. This approach substantially boosts performance on degraded images while maintain...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04780
• PDF: https://arxiv.org/pdf/2604.04780
• Project Page: https://haoxiangzhao12138.github.io/CLEAR/
• Github: https://github.com/haoxiangzhao12138/CLEAR

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Paper Espresso: From Paper Overload to Research Insight

📝 Summary:
Paper Espresso is an open-source LLM-powered platform that discovers, summarizes, and analyzes trending arXiv papers. It provides multi-granularity trend analysis, revealing AI research dynamics like a surge in RL for LLM reasoning and topic novelty correlating with community engagement.

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04562
• PDF: https://arxiv.org/pdf/2604.04562
• Project Page: https://mingzhe.space/assets/html/paper-espresso.html

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

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

#LLM #AIResearch #DataScience #ResearchTools #arXiv

✨POEMetric: The Last Stanza of Humanity

📝 Summary:
POEMetric evaluates LLM poetry generation across basic, creative, and quality dimensions, revealing significant gaps between human and machine capabilities in poetic expression. AI-generated summary L...

🔹 Publication Date: Published on Apr 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03695
• PDF: https://arxiv.org/pdf/2604.03695
• Github: https://github.com/Bingru-Li/POEMetric

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

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

#LLM #AIPoetry #AICreativity #NLP #HumanAI

✨ONE-SHOT: Compositional Human-Environment Video Synthesis via Spatial-Decoupled Motion Injection and Hybrid Context Integration

📝 Summary:
ONE-SHOT enables compositional human-environment video generation through disentangled signals, dynamic positional embeddings, and hybrid context integration for improved control and diversity. AI-gen...

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01043
• PDF: https://arxiv.org/pdf/2604.01043
• Project Page: https://martayang.github.io/ONE-SHOT/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨The Geometric Alignment Tax: Tokenization vs. Continuous Geometry in Scientific Foundation Models

📝 Summary:
Foundation models in biology and physics suffer from geometric distortion due to discrete categorical bottlenecks, with continuous objectives showing significantly better preservation of system geomet...

🔹 Publication Date: Published on Apr 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04155
• PDF: https://arxiv.org/pdf/2604.04155
• Github: https://github.com/prashantcraju/geometric-alignment-tax

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Emergent Compositional Communication for Latent World Properties

📝 Summary:
Multi-agent communication systems with Gumbel-Softmax emergently extract compositional representations of latent physical properties from video without supervision. This robust method supports planning and validates on real-world footage.

🔹 Publication Date: Published on Mar 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03266
• PDF: https://arxiv.org/pdf/2604.03266
• Github: https://github.com/TomekKaszynski/emergent-physics-comm

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Synthetic Sandbox for Training Machine Learning Engineering Agents

📝 Summary:
A multi-agent framework called SandMLE is introduced that generates synthetic machine learning engineering environments from limited seed tasks, enabling efficient on-policy reinforcement learning by ...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04872
• PDF: https://arxiv.org/pdf/2604.04872

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Cog-DRIFT: Exploration on Adaptively Reformulated Instances Enables Learning from Hard Reasoning Problems

📝 Summary:
Task reformulation and curriculum learning enable reinforcement learning from verifiable rewards to overcome exploration barriers in large language model post-training by transforming complex problems...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04767
• PDF: https://arxiv.org/pdf/2604.04767
• Github: https://github.com/dinobby/Cog-DRIFT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Do Audio-Visual Large Language Models Really See and Hear?

📝 Summary:
AVLLMs exhibit modality bias where visual representations dominate over audio cues during multimodal integration, despite audio semantics being present in intermediate layers. AI-generated summary Aud...

🔹 Publication Date: Published on Apr 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02605
• PDF: https://arxiv.org/pdf/2604.02605
• Project Page: https://ramaneswaran.github.io/avllm_interpretability/
• Github: https://github.com/ramaneswaran/avllm_interpretability

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models

📝 Summary:
Diffusion LLMs struggle with a quality-exploration dilemma; improving single-sample quality often limits reasoning path exploration. This paper explains why existing methods fail and proposes a new Independent Metropolis-Hastings sampler. This approach effectively balances quality and exploration...

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00375
• PDF: https://arxiv.org/pdf/2604.00375

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research