✨ISO-Bench: Can Coding Agents Optimize Real-World Inference Workloads?

📝 Summary:
ISO-Bench evaluates coding agents on real-world LLM inference optimization tasks using combined execution and LLM metrics. Agents often identify bottlenecks but fail to execute working solutions, highlighting that scaffolding is as important as the model itself.

🔹 Publication Date: Published on Feb 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.19594
• PDF: https://arxiv.org/pdf/2602.19594

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

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

#CodingAgents #LLMOptimization #AIResearch #Benchmarking #LargeLanguageModels

✨The Truthfulness Spectrum Hypothesis

📝 Summary:
This paper proposes the truthfulness spectrum hypothesis: LLMs contain truth directions ranging from domain-general to domain-specific. While general directions exist, domain-specific ones steer more effectively, with post-training reshaping this geometry to influence behaviors like sycophancy.

🔹 Publication Date: Published on Feb 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.20273
• PDF: https://arxiv.org/pdf/2602.20273
• Github: https://github.com/zfying/truth_spec

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

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

#LLMs #AIResearch #AIAlignment #NLP #Truthfulness

✨Intent Laundering: AI Safety Datasets Are Not What They Seem

📝 Summary:
AI safety datasets overrely on unrealistic triggering cues. This paper introduces intent laundering to remove these cues, revealing that models previously deemed safe become vulnerable. This method also works as a powerful jailbreaking technique, exposing a critical flaw in current AI safety eval...

🔹 Publication Date: Published on Feb 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.16729
• PDF: https://arxiv.org/pdf/2602.16729

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

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

#AISafety #JailbreakingAI #LLMSecurity #AIDatasets #AIEvaluation

✨The Trinity of Consistency as a Defining Principle for General World Models

📝 Summary:
This paper proposes the Trinity of Consistency modal, spatial, temporal as a foundational theoretical framework for General World Models. It systematically reviews multimodal learning through this lens and introduces CoW-Bench, a new benchmark for evaluating current and future models.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23152
• PDF: https://arxiv.org/pdf/2602.23152
• Project Page: https://openraiser.github.io/CoW-Bench/
• Github: https://github.com/openraiser/awesome-world-model-evolution

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OmniGAIA: Towards Native Omni-Modal AI Agents

📝 Summary:
OmniGAIA benchmark evaluates multi-modal agents on complex reasoning tasks across video, audio, and image modalities, while OmniAtlas agent improves tool-use capabilities through hindsight-guided tree...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22897
• PDF: https://arxiv.org/pdf/2602.22897
• Github: https://github.com/RUC-NLPIR/OmniGAIA

✨ Datasets citing this paper:
• https://huggingface.co/datasets/RUC-NLPIR/OmniGAIA
• https://huggingface.co/datasets/RUC-NLPIR/Omnimodal-Agent-SFT-2K

✨ Spaces citing this paper:
• https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving

📝 Summary:
A risk-aware framework for autonomous driving that uses world modeling and risk evaluation to generalize beyond expert demonstrations without requiring explicit expert supervision. AI-generated summar...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23259
• PDF: https://arxiv.org/pdf/2602.23259

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DyaDiT: A Multi-Modal Diffusion Transformer for Socially Favorable Dyadic Gesture Generation

📝 Summary:
DyaDiT is a multi-modal diffusion transformer that generates contextually appropriate human motion from dyadic audio signals by capturing interaction dynamics between two speakers. AI-generated summar...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23165
• PDF: https://arxiv.org/pdf/2602.23165

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨GeoWorld: Geometric World Models

📝 Summary:
GeoWorld addresses limitations in energy-based predictive world models by utilizing hyperbolic geometry to preserve latent state structures and improve long-horizon prediction performance. AI-generate...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23058
• PDF: https://arxiv.org/pdf/2602.23058
• Project Page: https://steve-zeyu-zhang.github.io/GeoWorld

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨veScale-FSDP: Flexible and High-Performance FSDP at Scale

📝 Summary:
veScale-FSDP introduces a redesigned fully sharded data parallel system with flexible sharding and structure-aware planning to improve scalability and efficiency for large-scale model training. AI-gen...

🔹 Publication Date: Published on Feb 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22437
• PDF: https://arxiv.org/pdf/2602.22437
• Github: https://github.com/volcengine/veScale

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Imagination Helps Visual Reasoning, But Not Yet in Latent Space

📝 Summary:
Research reveals that latent visual reasoning in multimodal models suffers from input-latent and latent-answer disconnects, leading to the proposal of CapImagine, a text-based approach that outperform...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22766
• PDF: https://arxiv.org/pdf/2602.22766
• Github: https://github.com/Michael4933/CapImagine

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Exploratory Memory-Augmented LLM Agent via Hybrid On- and Off-Policy Optimization

📝 Summary:
EMPO² is a hybrid reinforcement learning framework that enhances exploration for large language model agents by integrating memory mechanisms with on- and off-policy updates, demonstrating improved pe...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23008
• PDF: https://arxiv.org/pdf/2602.23008

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Causal Motion Diffusion Models for Autoregressive Motion Generation

📝 Summary:
Causal Motion Diffusion Models introduce a unified framework for autoregressive motion generation using a causal diffusion transformer in a semantically aligned latent space, enabling fast, high-quali...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22594
• PDF: https://arxiv.org/pdf/2602.22594
• Project Page: https://yu1ut.com/CMDM-HP/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Efficient Continual Learning in Language Models via Thalamically Routed Cortical Columns

📝 Summary:
TRC2 introduces a sparse, chunk-parallel architecture for language models to address continual learning challenges. It enables rapid adaptation and prevents catastrophic forgetting, improving the stability-plasticity tradeoff with efficient compute.

🔹 Publication Date: Published on Feb 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22479
• PDF: https://arxiv.org/pdf/2602.22479
• Project Page: https://trc2lm.github.io

🔹 Models citing this paper:
• https://huggingface.co/akhadangi/trc2

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AI Gamestore: Scalable, Open-Ended Evaluation of Machine General Intelligence with Human Games

📝 Summary:
AI systems were evaluated across a diverse set of human-designed games to assess general intelligence, revealing significant gaps in performance compared to human players, particularly in complex cogn...

🔹 Publication Date: Published on Feb 19

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.17594
• PDF: https://arxiv.org/pdf/2602.17594
• Project Page: https://aigamestore.org

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Accelerating Diffusion via Hybrid Data-Pipeline Parallelism Based on Conditional Guidance Scheduling

📝 Summary:
A hybrid parallelism framework for diffusion models that combines condition-based partitioning and adaptive pipeline scheduling to reduce inference latency while maintaining image quality across diffe...

🔹 Publication Date: Published on Feb 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.21760
• PDF: https://arxiv.org/pdf/2602.21760
• Github: https://github.com/kaist-dmlab/Hybridiff

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨From Blind Spots to Gains: Diagnostic-Driven Iterative Training for Large Multimodal Models

📝 Summary:
Diagnostic-driven Progressive Evolution enables continuous improvement of large multimodal models through iterative diagnosis and targeted data generation guided by identified weaknesses. AI-generated...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22859
• PDF: https://arxiv.org/pdf/2602.22859
• Github: https://github.com/hongruijia/DPE

🔹 Models citing this paper:
• https://huggingface.co/hongruijia/Qwen3_VL_8B_Instruct_DPE_v3
• https://huggingface.co/hongruijia/Qwen2.5-VL-7B-Instruct_DPE_v3
• https://huggingface.co/hongruijia/Qwen3_VL_8B_Instruct_DPE_v1

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AgentDropoutV2: Optimizing Information Flow in Multi-Agent Systems via Test-Time Rectify-or-Reject Pruning

📝 Summary:
AgentDropoutV2 is a test-time framework that optimizes multi-agent system information flow without retraining. It corrects errors and prunes irreparable agent outputs to prevent error propagation. This approach significantly boosts task performance and offers robust generalization and adaptivity.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23258
• PDF: https://arxiv.org/pdf/2602.23258
• Github: https://github.com/TonySY2/AgentDropoutV2

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

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

#MultiAgentSystems #AIResearch #InformationFlow #TestTimePruning #RobustAI

✨Search More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization

📝 Summary:
SMTL improves long-horizon agentic search by replacing sequential reasoning with parallel evidence acquisition. This framework achieves state-of-the-art performance and reduces reasoning steps by over 70% across diverse benchmarks, addressing efficiency and generalization challenges.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.22675
• PDF: https://arxiv.org/pdf/2602.22675

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

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

#AgenticSearch #AIResearch #Efficiency #Generalization #MachineLearning

✨MediX-R1: Open Ended Medical Reinforcement Learning

📝 Summary:
MediX-R1 is an open-ended reinforcement learning framework for medical multimodal LLMs. It uses diverse reward signals and LLM-based evaluation to enable clinically grounded, free-form answers, significantly improving reasoning on open-ended tasks.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23363
• PDF: https://arxiv.org/pdf/2602.23363
• Project Page: https://medix.cvmbzuai.com/
• Github: https://github.com/mbzuai-oryx/MediX-R1

🔹 Models citing this paper:
• https://huggingface.co/MBZUAI/MediX-R1-2B
• https://huggingface.co/MBZUAI/MediX-R1-8B
• https://huggingface.co/MBZUAI/MediX-R1-30B

✨ Datasets citing this paper:
• https://huggingface.co/datasets/MBZUAI/medix-rl-data

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

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

#MedicalAI #ReinforcementLearning #LLMs #MultimodalAI #AIResearch

✨EmbodMocap: In-the-Wild 4D Human-Scene Reconstruction for Embodied Agents

📝 Summary:
EmbodMocap is a dual-iPhone system for in-the-wild 4D human-scene reconstruction. It unifies human and scene data in a metric world frame, improving accuracy. This supports embodied AI tasks like animation and robot control.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23205
• PDF: https://arxiv.org/pdf/2602.23205
• Project Page: https://wenjiawang0312.github.io/projects/embodmocap/
• Github: https://github.com/WenjiaWang0312/EmbodMocap

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

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

#EmbodiedAI #4DReconstruction #ComputerVision #Robotics #Animation

✨Echoes Over Time: Unlocking Length Generalization in Video-to-Audio Generation Models

📝 Summary:
MMHNet uses hierarchical methods and non-causal Mamba for video-to-audio generation. It achieves length generalization, allowing training on short videos to generate over 5 minutes of high-quality audio, outperforming prior models.

🔹 Publication Date: Published on Feb 24

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.20981
• PDF: https://arxiv.org/pdf/2602.20981

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

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

#VideoToAudio #LengthGeneralization #Mamba #DeepLearning #AIResearch