✨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

✨Type-Checked Compliance: Deterministic Guardrails for Agentic Financial Systems Using Lean 4 Theorem Proving

📝 Summary:
The Lean-Agent Protocol ensures deterministic regulatory compliance for financial AI. It uses Lean 4 theorem proving to auto-formalize policies, verifying agent actions as mathematical conjectures for cryptographic-level certainty, addressing LLM probabilistic nature.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01483
• PDF: https://arxiv.org/pdf/2604.01483
• Project Page: https://axiom.devrashie.space
• Github: https://github.com/arkanemystic/lean-agent-protocol

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

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

#FormalVerification #AICompliance #FinTech #Lean4 #LLMAgents

✨Scaling Teams or Scaling Time? Memory Enabled Lifelong Learning in LLM Multi-Agent Systems

📝 Summary:
This paper introduces LLMA-Mem, a memory framework for LLM multi-agent systems. It finds that scaling is non-monotonic; optimized experience reuse allows smaller teams to outperform larger ones, improving long-term performance and reducing cost.

🔹 Publication Date: Published on Mar 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03295
• PDF: https://arxiv.org/pdf/2604.03295
• Github: https://github.com/ShanglinWu/MAS_lifelong_learning

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨BidirLM: From Text to Omnimodal Bidirectional Encoders by Adapting and Composing Causal LLMs

📝 Summary:
BidirLM adapts causal LLMs into bidirectional encoders, overcoming catastrophic forgetting and integrating specialized models. It employs a prior masking phase, weight merging, and data mixture, outperforming alternatives on text, vision, and audio benchmarks.

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02045
• PDF: https://arxiv.org/pdf/2604.02045

🔹 Models citing this paper:
• https://huggingface.co/BidirLM/BidirLM-Omni-2.5B-Embedding
• https://huggingface.co/BidirLM/BidirLM-0.6B-Embedding
• https://huggingface.co/BidirLM/BidirLM-1.7B-Embedding

✨ Datasets citing this paper:
• https://huggingface.co/datasets/BidirLM/BidirLM-Contrastive

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

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

#LLM #MultimodalAI #DeepLearning #AIResearch #ModelAdaptation

✨Beyond Accuracy: Unveiling Inefficiency Patterns in Tool-Integrated Reasoning

📝 Summary:
The paper introduces PTE Prefill Token Equivalents, a hardware-aware metric for Tool-Integrated Reasoning efficiency. PTE better measures real inference latency than token counts by accounting for KV-Cache inefficiencies and long tool responses. Higher PTE costs often indicate lower reasoning cor...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.05404
• PDF: https://arxiv.org/pdf/2604.05404
• Github: https://github.com/sqs-ustc/tool-reasoning-framework-PTE

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨FactReview: Evidence-Grounded Reviews with Literature Positioning and Execution-Based Claim Verification

📝 Summary:
FactReview is an evidence-grounded peer review system for machine learning that analyzes manuscript claims through claim extraction, literature positioning, and execution-based verification to provide...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04074
• PDF: https://arxiv.org/pdf/2604.04074

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video Understanding

📝 Summary:
Video-MME-v2 presents a comprehensive benchmark for evaluating video understanding models through a progressive hierarchy and group-based evaluation to assess robustness and faithfulness. AI-generated...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2604.05015
• PDF: https://arxiv.org/pdf/2604.05015
• Project Page: https://video-mme-v2.netlify.app/
• Github: https://github.com/MME-Benchmarks/Video-MME-v2

✨ Datasets citing this paper:
• https://huggingface.co/datasets/MME-Benchmarks/Video-MME-v2

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learning to Retrieve from Agent Trajectories

📝 Summary:
Retrieval models for agentic search should be trained directly from agent interaction data using a new paradigm that mines supervision from multi-step agent trajectories and incorporates relevance int...

🔹 Publication Date: Published on Mar 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04949
• PDF: https://arxiv.org/pdf/2604.04949
• Project Page: https://yuqi-zhou.github.io/LRAT-homepage/
• Github: https://github.com/Yuqi-Zhou/LRAT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Claw-Eval: Toward Trustworthy Evaluation of Autonomous Agents

📝 Summary:
Claw-Eval addresses limitations in agent benchmarks by providing comprehensive evaluation across multiple modalities with trajectory-aware grading and safety assessments. AI-generated summary Large la...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.06132
• PDF: https://arxiv.org/pdf/2604.06132
• Project Page: https://claw-eval.github.io/
• Github: https://github.com/claw-eval/claw-eval

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Vanast: Virtual Try-On with Human Image Animation via Synthetic Triplet Supervision

📝 Summary:
Vanast is a unified framework that generates garment-transferred human animation videos by combining image-based virtual try-on and pose-driven animation in a single process, addressing issues like id...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04934
• PDF: https://arxiv.org/pdf/2604.04934
• Project Page: https://hyunsoocha.github.io/vanast/
• Github: https://github.com/snuvclab/vanast

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ACES: Who Tests the Tests? Leave-One-Out AUC Consistency for Code Generation

📝 Summary:
Researchers address the challenge of selecting correct code candidates from LLM-generated outputs by developing ACES, a method that ranks tests based on their ability to distinguish correct from incor...

🔹 Publication Date: Published on Apr 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03922
• PDF: https://arxiv.org/pdf/2604.03922

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MMEmb-R1: Reasoning-Enhanced Multimodal Embedding with Pair-Aware Selection and Adaptive Control

📝 Summary:
An adaptive multimodal embedding framework that selectively applies reasoning through latent variables and reinforcement learning to improve efficiency and performance on benchmark tasks. AI-generated...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.06156
• PDF: https://arxiv.org/pdf/2604.06156

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨In-Place Test-Time Training

📝 Summary:
In-Place Test-Time Training enables large language models to adapt parameters during inference by modifying the final projection matrix in MLP blocks with a task-aligned objective and efficient update...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.06169
• PDF: https://arxiv.org/pdf/2604.06169
• Github: https://github.com/ByteDance-Seed/In-Place-TTT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Demystifying When Pruning Works via Representation Hierarchies

📝 Summary:
Network pruning affects different representation spaces differently, leading to varying performance across tasks due to instability in probability space transformations during generation. AI-generated...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.24652
• PDF: https://arxiv.org/pdf/2603.24652
• Github: https://github.com/CASE-Lab-UMD/Pruning-on-Representations

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Experience Transfer for Multimodal LLM Agents in Minecraft Game

📝 Summary:
Echo is a transfer-oriented memory framework for multimodal LLM agents that enables efficient task solving in complex game environments by deriving actionable knowledge from prior interactions through...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.05533
• PDF: https://arxiv.org/pdf/2604.05533

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Action Images: End-to-End Policy Learning via Multiview Video Generation

📝 Summary:
World action models that formulate policy learning as multiview video generation use pixel-grounded action images to enable zero-shot policy learning without separate action modules. AI-generated summ...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.06168
• PDF: https://arxiv.org/pdf/2604.06168
• Project Page: https://actionimages.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Scientific Graphics Program Synthesis via Dual Self-Consistency Reinforcement Learning

📝 Summary:
A closed-loop framework for graphics program synthesis combines a large-scale dataset and benchmark with a novel reinforcement learning optimization method to improve the generation of executable TikZ...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.06079
• PDF: https://arxiv.org/pdf/2604.06079

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MedGemma 1.5 Technical Report

📝 Summary:
MedGemma 1.5 4B enhances medical AI capabilities through expanded multimodal support and improved performance across medical imaging, document understanding, and clinical reasoning tasks. AI-generated...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.05081
• PDF: https://arxiv.org/pdf/2604.05081

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ThinkTwice: Jointly Optimizing Large Language Models for Reasoning and Self-Refinement

📝 Summary:
ThinkTwice is a two-phase framework that jointly optimizes large language models for reasoning and self-refinement using Group Relative Policy Optimization, demonstrating improved performance on mathe...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01591
• PDF: https://arxiv.org/pdf/2604.01591
• Github: https://github.com/CSSLab/ThinkTwice

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research