✨InfoPO: Information-Driven Policy Optimization for User-Centric Agents

📝 Summary:
InfoPO optimizes agent-user collaboration for underspecified requests. It uses an information-gain reward to credit valuable turns that reduce uncertainty, improving decision-making and outperforming multi-turn RL baselines.

🔹 Publication Date: Published on Feb 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00656
• PDF: https://arxiv.org/pdf/2603.00656
• Github: https://github.com/kfq20/InfoPO

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

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

#ReinforcementLearning #AI #HumanComputerInteraction #InformationTheory #AIagents

✨Chain of World: World Model Thinking in Latent Motion

📝 Summary:
CoWVLA unifies world-model temporal reasoning with disentangled latent motion representation to improve visuomotor learning efficiency. This new approach overcomes limitations of existing VLA models and outperforms them on robotic simulation benchmarks.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03195
• PDF: https://arxiv.org/pdf/2603.03195
• Project Page: https://fx-hit.github.io/cowvla-io/
• Github: https://fx-hit.github.io/cowvla-io/

🔹 Models citing this paper:
• https://huggingface.co/hitfx/CoWVLA

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

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

#WorldModels #Robotics #MachineLearning #VisuomotorLearning #DeepLearning

✨Surgical Post-Training: Cutting Errors, Keeping Knowledge

📝 Summary:
Surgical Post-Training SPoT efficiently improves LLM reasoning while preventing catastrophic forgetting. It employs data rectification with an Oracle and a novel binary cross-entropy objective. SPoT enhanced Qwen3-8B accuracy by 6.2 percent using minimal data and training time.

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01683
• PDF: https://arxiv.org/pdf/2603.01683
• Github: https://github.com/Visual-AI/SPoT

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

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

#LLM #CatastrophicForgetting #MachineLearning #AI #DeepLearning

✨Whisper-RIR-Mega: A Paired Clean-Reverberant Speech Benchmark for ASR Robustness to Room Acoustics

📝 Summary:
Whisper-RIR-Mega dataset evaluates ASR model robustness to reverberation by pairing clean and reverberant speech samples with stratified splits based on RT60 and DRR metrics. AI-generated summary We i...

🔹 Publication Date: Published on Feb 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02252
• PDF: https://arxiv.org/pdf/2603.02252
• Project Page: https://huggingface.co/datasets/mandipgoswami/whisper-rirmega-bench
• Github: https://github.com/mandip42/whisper-rirmega-bench

✨ Datasets citing this paper:
• https://huggingface.co/datasets/mandipgoswami/whisper-rirmega-bench

✨ Spaces citing this paper:
• https://huggingface.co/spaces/mandipgoswami/whisper-rirmega-benchmark

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learning When to Act or Refuse: Guarding Agentic Reasoning Models for Safe Multi-Step Tool Use

📝 Summary:
MOSAIC is a framework aligning agentic models for safe multi-step tool use, employing explicit safety reasoning and refusal. It significantly reduces harmful actions, increases refusal for unsafe tasks, cuts privacy leakage, and preserves benign performance.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03205
• PDF: https://arxiv.org/pdf/2603.03205
• Project Page: https://aradhye2002.github.io/mosaic-agent-safety/

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

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

#AISafety #AIAgents #ResponsibleAI #LLMs #AIAlignment

✨Spilled Energy in Large Language Models

📝 Summary:
Reinterpreting LLM softmax as an Energy-Based Model enables training-free hallucination detection. New energy metrics from output logits identify errors and biases without training overhead, demonstrating robust cross-task generalization.

🔹 Publication Date: Published on Feb 21

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.18671
• PDF: https://arxiv.org/pdf/2602.18671
• Github: https://github.com/OmnAI-Lab/spilled-energy

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

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

#LLM #EnergyBasedModels #HallucinationDetection #AISafety #ArtificialIntelligence

✨Towards Simulating Social Media Users with LLMs: Evaluating the Operational Validity of Conditioned Comment Prediction

📝 Summary:
CCP evaluates LLMs simulating social media users. Supervised fine-tuning improves text structure but degrades semantic accuracy, as models infer from behavioral histories without explicit conditioning. Prioritize authentic behavioral traces.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22752
• PDF: https://arxiv.org/pdf/2602.22752
• Project Page: https://nsschw.github.io/Turing-TWONy/
• Github: https://github.com/nsschw/Conditioned-Comment-Prediction

🔹 Models citing this paper:
• https://huggingface.co/nsschw/echo-Llama-3.1-8B-Instruct-eng
• https://huggingface.co/nsschw/echo-Llama-3.1-8B-Instruct-ger

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

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

#LLMs #SocialMedia #AISimulation #NLP #AIResearch

✨Conditioned Activation Transport for T2I Safety Steering

📝 Summary:
Current T2I models generate unsafe content, and linear steering degrades image quality. This paper proposes Conditioned Activation Transport CAT, which uses geometric conditioning and nonlinear transport maps to activate only in unsafe regions. CAT significantly reduces unsafe content generation ...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03163
• PDF: https://arxiv.org/pdf/2603.03163
• Github: https://github.com/NASK-AISafety/conditional-activation-transport

✨ Datasets citing this paper:
• https://huggingface.co/datasets/NASK-PIB/SafeSteerDataset

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

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

#AISafety #TextToImage #GenerativeAI #DeepLearning #AIethics

✨Transform-Invariant Generative Ray Path Sampling for Efficient Radio Propagation Modeling

📝 Summary:
A machine learning framework using generative flow networks with experience replay, uniform exploration, and physics-based masking enables fast and accurate radio propagation path sampling with signif...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01655
• PDF: https://arxiv.org/pdf/2603.01655
• Project Page: https://differt.rtfd.io/npjwt2026/notebooks/sampling-paths.html
• Github: https://github.com/jeertmans/sampling-paths

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MNN: A Universal and Efficient Inference Engine

📝 Summary:
MNN is an efficient deep learning inference engine for mobile devices. It addresses compatibility and resource limits through pre-inference, kernel optimization, and backend abstraction, outperforming other lightweight frameworks.

🔹 Publication Date: Published on Feb 27, 2020

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2002.12418
• PDF: https://arxiv.org/pdf/2002.12418
• Github: https://github.com/alibaba/MNN

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

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

#DeepLearning #MobileAI #EdgeAI #Optimization #MachineLearning

✨CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance

📝 Summary:
This paper reinterprets Classifier-Free Guidance CFG as a control system for diffusion models. It introduces Sliding Mode Control CFG SMC-CFG to overcome instability in existing linear CFG methods. SMC-CFG improves semantic alignment and stability across various guidance scales.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03281
• PDF: https://arxiv.org/pdf/2603.03281
• Project Page: https://hanyang-21.github.io/CFG-Ctrl
• Github: https://github.com/hanyang-21/CFG-Ctrl

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

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

#DiffusionModels #GenerativeAI #ControlSystems #MachineLearning #AIResearch

✨Multi-Domain Riemannian Graph Gluing for Building Graph Foundation Models

📝 Summary:
This paper proposes GraphGlue, a framework that uses Riemannian geometry and neural manifold gluing to integrate knowledge from diverse graph domains. It merges datasets into a unified manifold for systematic understanding of knowledge transfer. GraphGlue achieves superior performance and shows t...

🔹 Publication Date: Published on Feb 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00618
• PDF: https://arxiv.org/pdf/2603.00618
• Github: https://github.com/RiemannGraph/GraphGlue

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

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

#GraphFoundationModels #RiemannianGeometry #GraphAI #KnowledgeTransfer #MachineLearning

✨Next Embedding Prediction Makes World Models Stronger

📝 Summary:
NE-Dreamer uses a temporal transformer to predict next-step encoder embeddings, enabling strong model-based reinforcement learning without decoders. This approach learns coherent state representations and achieves strong performance on DeepMind Control Suite and challenging DMLab tasks.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02765
• PDF: https://arxiv.org/pdf/2603.02765
• Project Page: https://corl-team.github.io/nedreamer/
• Github: https://github.com/corl-team/nedreamer

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DynaMoE: Dynamic Token-Level Expert Activation with Layer-Wise Adaptive Capacity for Mixture-of-Experts Neural Networks

📝 Summary:
DynaMoE presents a dynamic Mixture-of-Experts framework that adapts expert activation and capacity allocation based on input complexity and task requirements, improving parameter efficiency and traini...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01697
• PDF: https://arxiv.org/pdf/2603.01697

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SciDER: Scientific Data-centric End-to-end Researcher

📝 Summary:
SciDER automates scientific research by processing raw experimental data through collaborative agents that generate hypotheses and experimental designs while executing code, demonstrating superior per...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01421
• PDF: https://arxiv.org/pdf/2603.01421
• Project Page: https://harryluumn.github.io/scider-proj-page/
• Github: https://github.com/leonardodalinky/SciDER

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨BBQ-to-Image: Numeric Bounding Box and Qolor Control in Large-Scale Text-to-Image Models

📝 Summary:
BBQ is a text-to-image model that enables precise numeric control over object attributes through structured-text conditioning without architectural changes. AI-generated summary Text-to-image models h...

🔹 Publication Date: Published on Feb 24

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.20672
• PDF: https://arxiv.org/pdf/2602.20672

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨GroupGPT: A Token-efficient and Privacy-preserving Agentic Framework for Multi-User Chat Assistant

📝 Summary:
GroupGPT is a token-efficient and privacy-preserving framework for multi-user chat assistance that uses a small-large model collaboration approach to improve intervention timing and response accuracy ...

🔹 Publication Date: Published on Mar 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01059
• PDF: https://arxiv.org/pdf/2603.01059
• Github: https://github.com/Eliot-Shen/GroupGPT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Transformers converge to invariant algorithmic cores

📝 Summary:
Independently trained transformers converge to shared low-dimensional algorithmic cores. These compact invariants reveal the computational essence across training runs and scales, suggesting a new focus for mechanistic interpretability.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22600
• PDF: https://arxiv.org/pdf/2602.22600

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Humans and LLMs Diverge on Probabilistic Inferences

📝 Summary:
LLMs consistently fail to replicate human probabilistic reasoning patterns in open-ended inferences, despite strong performance on logical and mathematical tasks. The ProbCOPA dataset reveals this key divergence, highlighting a need to evaluate AI reasoning beyond deterministic settings.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23546
• PDF: https://arxiv.org/pdf/2602.23546
• Project Page: https://grvkamath.github.io/probcopa-demo/index.html
• Github: https://github.com/McGill-NLP/probabilistic-reasoning

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Easy to Learn, Yet Hard to Forget: Towards Robust Unlearning Under Bias

📝 Summary:
Machine unlearning in biased models suffers from shortcut unlearning, where bias attributes are forgotten instead of class attributes. The CUPID framework addresses this by partitioning data based on loss sharpness and disentangling causal and bias pathways for targeted updates, achieving state-o...

🔹 Publication Date: Published on Feb 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.21773
• PDF: https://arxiv.org/pdf/2602.21773

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research