✨Reasoning Core: A Scalable Procedural Data Generation Suite for Symbolic Pre-training and Post-Training

📝 Summary:
Reasoning Core is a new scalable system that procedurally generates verifiable symbolic reasoning data across diverse formal domains. Mixing this data into pre-training improves language model reasoning abilities while preserving language modeling quality.

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02208
• PDF: https://arxiv.org/pdf/2603.02208
• Project Page: https://github.com/sileod/reasoning_core/
• Github: https://github.com/sileod/reasoning_core

✨ Datasets citing this paper:
• https://huggingface.co/datasets/reasoning-core/symbolic-pretraining-pile
• https://huggingface.co/datasets/reasoning-core/symbolic-reasoning-env

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

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

#AI #LLM #SymbolicReasoning #DataGeneration #MachineLearning

✨Spectral Attention Steering for Prompt Highlighting

📝 Summary:
SEKA and AdaSEKA introduce training-free attention steering by editing key embeddings using spectral decomposition. This amplifies attention for specific tokens, outperforming baselines with less memory and latency, compatible with optimized attention.

🔹 Publication Date: Published on Mar 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01281
• PDF: https://arxiv.org/pdf/2603.01281
• Github: https://github.com/waylonli/SEKA

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

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

#AttentionMechanisms #NLP #DeepLearning #MachineLearning #AI

✨OpenAutoNLU: Open Source AutoML Library for NLU

📝 Summary:
OpenAutoNLU is an open-source AutoML library for NLU tasks like text classification and named entity recognition. Its key innovation is data-aware training selection requiring no manual configuration. It also offers integrated diagnostics, out-of-distribution detection, and LLM features through a...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01824
• PDF: https://arxiv.org/pdf/2603.01824
• Project Page: https://openautonlu.dev
• Github: https://github.com/mts-ai/OpenAutoNLU

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

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

#AutoML #NLU #LLM #OpenSource #MachineLearning

✨SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale

📝 Summary:
SWE-rebench V2 presents a new language-agnostic automated pipeline to create a large-scale dataset of over 32,000 software engineering tasks across 20 languages and 3,600 repositories. It provides reproducible environments and reliable tests, validated by LLMs, to advance training for SWE agents.

🔹 Publication Date: Published on Feb 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23866
• PDF: https://arxiv.org/pdf/2602.23866
• Github: https://huggingface.co/collections/nebius/swe-rebench-v2

✨ Datasets citing this paper:
• https://huggingface.co/datasets/nebius/SWE-rebench-V2
• https://huggingface.co/datasets/nebius/SWE-rebench-V2-PRs

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

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

#SoftwareEngineering #LLMs #AI #Dataset #SWEAgents

✨Monocular Mesh Recovery and Body Measurement of Female Saanen Goats

📝 Summary:
This paper introduces a novel 3D body measurement system for Saanen goats. It uses a new parametric shape model and a multi-view RGBD dataset to enable accurate single-view 3D reconstruction and automated measurement of key body dimensions, improving precision livestock farming.

🔹 Publication Date: Published on Feb 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.19896
• PDF: https://arxiv.org/pdf/2602.19896
• Github: https://github.com/bojin-nwafu/Female-Saanen-Goats

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

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

#3DReconstruction #ComputerVision #PrecisionLivestock #AnimalScience #AgriTech

✨Efficient RLVR Training via Weighted Mutual Information Data Selection

📝 Summary:
InSight is a new data sampling method for RL training that improves efficiency. It considers both data difficulty and epistemic uncertainty, unlike prior methods. This Bayesian modeling approach achieves state-of-the-art performance and significantly accelerates training.

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01907
• PDF: https://arxiv.org/pdf/2603.01907

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

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

#ReinforcementLearning #MachineLearning #DataScience #BayesianModeling #AI

✨ProtegoFed: Backdoor-Free Federated Instruction Tuning with Interspersed Poisoned Data

📝 Summary:
ProtegoFed is a new federated instruction tuning framework. It detects and removes widespread poisoned data across clients using frequency domain gradient analysis and collaborative clustering, reducing attack success to almost zero while maintaining utility.

🔹 Publication Date: Published on Feb 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00516
• PDF: https://arxiv.org/pdf/2603.00516
• Project Page: https://github.com/dongdongzhaoUP/ProtegoFed
• Github: https://github.com/dongdongzhaoUP/ProtegoFed

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

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

#FederatedLearning #AIsecurity #DataPoisoning #MachineLearning #AIResearch

✨Using Songs to Improve Kazakh Automatic Speech Recognition

📝 Summary:
This study improves Kazakh ASR for low-resource languages by using songs as a novel data source. Fine-tuning models with song data, especially combined with existing corpora, significantly boosts performance and offers meaningful adaptation gains.

🔹 Publication Date: Published on Mar 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00961
• PDF: https://arxiv.org/pdf/2603.00961

✨ Datasets citing this paper:
• https://huggingface.co/datasets/yeshpanovrustem/kazakh_songs_asr

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

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

#KazakhASR #LowResourceNLP #SpeechRecognition #DataInnovation #MachineLearning

✨Synthetic Visual Genome 2: Extracting Large-scale Spatio-Temporal Scene Graphs from Videos

📝 Summary:
A large video scene graph dataset, SVG2, and a new model, TRaSER, are introduced. TRaSER generates spatio-temporal scene graphs, significantly improving relation, object, and attribute prediction, and boosting video question answering accuracy.

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23543
• PDF: https://arxiv.org/pdf/2602.23543
• Project Page: https://uwgzq.github.io/papers/SVG2/

🔹 Models citing this paper:
• https://huggingface.co/UWGZQ/TRASER

✨ Datasets citing this paper:
• https://huggingface.co/datasets/UWGZQ/Synthetic_Visual_Genome2

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

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

#VideoSceneGraphs #SpatioTemporal #ComputerVision #VideoQA #DeepLearning

✨PhotoBench: Beyond Visual Matching Towards Personalized Intent-Driven Photo Retrieval

📝 Summary:
PhotoBench introduces a new benchmark for personalized, intent-driven photo retrieval from authentic albums, moving beyond visual matching. It shows current models struggle with non-visual constraints and multi-source fusion, stressing the need for robust agentic reasoning systems.

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01493
• PDF: https://arxiv.org/pdf/2603.01493
• Github: https://github.com/LaVieEnRose365/PhotoBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Unified Vision-Language Modeling via Concept Space Alignment

📝 Summary:
V-SONAR extends the text-only SONAR embedding space to support vision-language tasks through post-hoc alignment, enabling zero-shot visual concept understanding and outperforming state-of-the-art mode...

🔹 Publication Date: Published on Mar 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01096
• PDF: https://arxiv.org/pdf/2603.01096

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Classroom Final Exam: An Instructor-Tested Reasoning Benchmark

📝 Summary:
Classroom Final Exam CFE is a multimodal benchmark using authentic university STEM exam problems to assess LLM reasoning. Frontier models achieve only ~60% accuracy, struggling with multi-step solutions and maintaining intermediate states. This highlights significant room for improvement.

🔹 Publication Date: Published on Feb 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.19517
• PDF: https://arxiv.org/pdf/2602.19517
• Project Page: https://analogyai.ai/cfe_bench.html
• Github: https://github.com/Analogy-AI/CFE_Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Cryo-Bench: Benchmarking Foundation Models for Cryosphere Applications

📝 Summary:
Cryo-Bench benchmarks Geo-Foundation Models GFMs for cryosphere tasks, addressing a data gap. It evaluates 14 GFMs, finding they adapt well despite limited pretraining. For optimal results, encoder fine-tuning with hyperparameter optimization is recommended.

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01576
• PDF: https://arxiv.org/pdf/2603.01576
• Github: https://github.com/Sk-2103/Cryo-Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Planning from Observation and Interaction

📝 Summary:
This paper presents a planning-based Inverse Reinforcement Learning algorithm for real-world robot manipulation. It learns effectively from observation and interaction alone, without prior knowledge or pre-training. The approach demonstrates superior sample efficiency and enables online transfer ...

🔹 Publication Date: Published on Feb 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.24121
• PDF: https://arxiv.org/pdf/2602.24121
• Project Page: https://uwrobotlearning.github.io/mpail2/
• Github: https://github.com/UWRobotLearning/mpail2

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨CMI-RewardBench: Evaluating Music Reward Models with Compositional Multimodal Instruction

📝 Summary:
CMI-RewardBench establishes a comprehensive ecosystem for evaluating music reward models under compositional multimodal instruction. It provides large-scale datasets, a unified benchmark for various alignment tasks, and CMI reward models that correlate strongly with human judgments.

🔹 Publication Date: Published on Feb 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00610
• PDF: https://arxiv.org/pdf/2603.00610
• Github: https://github.com/Haiwen-Xia/CMI-RewardBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities

📝 Summary:
SteerEval is a new hierarchical benchmark to evaluate large language model controllability across language, sentiment, and personality. It shows that control often degrades at finer-grained levels, providing a framework for safer and more controllable LLM behavior.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02578
• PDF: https://arxiv.org/pdf/2603.02578

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Beyond Length Scaling: Synergizing Breadth and Depth for Generative Reward Models

📝 Summary:
Generative Reward Models can be improved by structuring Chain-of-Thought reasoning into breadth and depth components and optimizing them through supervised fine-tuning and reinforcement learning with ...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01571
• PDF: https://arxiv.org/pdf/2603.01571
• Project Page: https://huggingface.co/collections/DonJoey/mix-grm

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DREAM: Where Visual Understanding Meets Text-to-Image Generation

📝 Summary:
DREAM is a unified multimodal framework that combines visual representation learning and text-to-image generation through progressive masking and semantically aligned decoding, achieving superior perf...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02667
• PDF: https://arxiv.org/pdf/2603.02667
• Github: https://github.com/chaoli-charlie/dream

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨PRISM: Pushing the Frontier of Deep Think via Process Reward Model-Guided Inference

📝 Summary:
PRISM is a Process Reward Model-guided inference algorithm that enhances DEEPTHINK systems by using step-level verification to improve population refinement and solution aggregation, achieving strong ...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02479
• PDF: https://arxiv.org/pdf/2603.02479

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨UniG2U-Bench: Do Unified Models Advance Multimodal Understanding?

📝 Summary:
Unified multimodal models generally underperform specialized VLMs in generation-to-understanding tasks. However, they show consistent enhancements in spatial intelligence, visual illusions, and multi-round reasoning. This highlights the need for diverse training data to unlock their full potential.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03241
• PDF: https://arxiv.org/pdf/2603.03241
• Project Page: https://nssmd.github.io/unig2u.github.io/
• Github: https://github.com/nssmd/UniG2U

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research