✨Vision Transformer Finetuning Benefits from Non-Smooth Components

📝 Summary:
Vision transformer components exhibit varying plasticity levels that correlate with finetuning performance, challenging the assumption that smoothness is always beneficial. AI-generated summary The sm...

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06883
• PDF: https://arxiv.org/pdf/2602.06883
• Github: https://github.com/ambroiseodt/vit-plasticity

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders

📝 Summary:
AudioSAE applies sparse autoencoders to Whisper and HuBERT models, extracting stable acoustic and semantic features. These features disentangle information, reduce false speech detections, and correlate with human EEG, demonstrating practical utility.

🔹 Publication Date: Published on Feb 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.05027
• PDF: https://arxiv.org/pdf/2602.05027

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

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

#AudioAI #SparseAutoencoders #MachineLearning #SpeechRecognition #Neuroscience

✨Group-Evolving Agents: Open-Ended Self-Improvement via Experience Sharing

📝 Summary:
Group-Evolving Agents GEA enable open-ended self-improvement by treating agent groups as evolutionary units, allowing efficient experience sharing. GEA significantly outperforms state-of-the-art self-evolving methods on coding benchmarks, demonstrating enhanced robustness and sustained progress.

🔹 Publication Date: Published on Feb 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.04837
• PDF: https://arxiv.org/pdf/2602.04837

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

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

#AI #EvolutionaryAI #MultiAgentSystems #OpenEndedLearning #MachineLearning

✨Urban Spatio-Temporal Foundation Models for Climate-Resilient Housing: Scaling Diffusion Transformers for Disaster Risk Prediction

📝 Summary:
This paper presents Skjold-DiT, a diffusion-transformer framework predicting building-level climate risks. It integrates urban data and transportation networks to generate accessibility layers for emergency response and intelligent vehicles. Experiments evaluate its prediction quality and cross-c...

🔹 Publication Date: Published on Feb 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06129
• PDF: https://arxiv.org/pdf/2602.06129

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

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

#UrbanAI #ClimateResilience #DisasterRisk #DiffusionModels #SpatioTemporalAI

✨Canzona: A Unified, Asynchronous, and Load-Balanced Framework for Distributed Matrix-based Optimizers

📝 Summary:
Canzona presents a unified asynchronous framework that addresses the conflict between matrix-based optimizers and distributed tensor fragmentation in LLM training, improving efficiency and reducing la...

🔹 Publication Date: Published on Feb 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06079
• PDF: https://arxiv.org/pdf/2602.06079

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨QuantLRM: Quantization of Large Reasoning Models via Fine-Tuning Signals

📝 Summary:
QuantLRM improves Large Reasoning Model quantization by using weight update magnitudes from fine-tuning to estimate channel importance. It protects both smallest and largest updates, consistently outperforming traditional methods and applying even to non-fine-tuned models.

🔹 Publication Date: Published on Jan 31

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02581
• PDF: https://arxiv.org/pdf/2602.02581
• Github: https://github.com/psunlpgroup/QuantLRM

🔹 Models citing this paper:
• https://huggingface.co/nanzhang/QuantLRM-R1-Qwen-32B-3-bit
• https://huggingface.co/nanzhang/QuantLRM-R1-Llama-70B-3-bit
• https://huggingface.co/nanzhang/QuantLRM-R1-Qwen3-8B-3-bit

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

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

#Quantization #LargeLanguageModels #DeepLearning #AI #ModelCompression

✨Table-as-Search: Formulate Long-Horizon Agentic Information Seeking as Table Completion

📝 Summary:
Table-as-Search TaS reformulates information seeking as table completion to robustly manage long-horizon search states. By mapping queries to structured tables, TaS explicitly tracks progress and plans, significantly outperforming baselines in complex search tasks.

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06724
• PDF: https://arxiv.org/pdf/2602.06724
• Github: https://github.com/AIDC-AI/Marco-DeepResearch/

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

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

#AI #InformationRetrieval #AgenticAI #TableCompletion #SearchAlgorithms

✨OmniVideo-R1: Reinforcing Audio-visual Reasoning with Query Intention and Modality Attention

📝 Summary:
OmniVideo-R1 is a reinforced framework that enhances audio-visual understanding. It uses self-supervised query-intensive grounding and contrastive modality-attentive fusion. Experiments show OmniVideo-R1 consistently outperforms baselines, demonstrating its effectiveness.

🔹 Publication Date: Published on Feb 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.05847
• PDF: https://arxiv.org/pdf/2602.05847

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

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

#AudioVisualAI #SelfSupervisedLearning #DeepLearning #MultimodalAI #AIResearch

✨SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue

📝 Summary:
SEAD enables service dialogue agents to learn effective strategies through self-evolving, decoupled user modeling. This trains agents without large human annotations, significantly improving task completion and dialogue efficiency compared to existing models.

🔹 Publication Date: Published on Feb 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.03548
• PDF: https://arxiv.org/pdf/2602.03548
• Github: https://github.com/Da1yuqin/SEAD

🔹 Models citing this paper:
• https://huggingface.co/dayll/SEAD-14B

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

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

#AI #ConversationalAI #ReinforcementLearning #NLP #AIagents

✨ReMiT: RL-Guided Mid-Training for Iterative LLM Evolution

📝 Summary:
ReMiT introduces a bidirectional training approach for LLMs. It leverages RL-guided mid-training to dynamically reweight tokens, improving pre-training performance and sustaining gains throughout post-training. This creates a self-reinforcing, iterative evolution cycle for LLMs.

🔹 Publication Date: Published on Feb 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.03075
• PDF: https://arxiv.org/pdf/2602.03075

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

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

#LLM #ReinforcementLearning #MachineLearning #AITraining #DeepLearning

✨Self-Improving World Modelling with Latent Actions

📝 Summary:
SWIRL learns world models from state-only data by treating actions as latent variables. It alternates forward and inverse dynamics modeling, using information maximization and ELBO, to achieve improved performance across diverse reasoning and planning tasks.

🔹 Publication Date: Published on Feb 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06130
• PDF: https://arxiv.org/pdf/2602.06130

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

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

#WorldModels #ReinforcementLearning #LatentVariables #MachineLearning #AI

✨Pisets: A Robust Speech Recognition System for Lectures and Interviews

📝 Summary:
Pisets is a robust Russian speech-to-text system combining Wav2Vec2, AST, and Whisper models. It uses curriculum learning and uncertainty modeling to improve accuracy and reduce hallucinations for long audio, outperforming other Whisper variants.

🔹 Publication Date: Published on Jan 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.18415
• PDF: https://arxiv.org/pdf/2601.18415

🔹 Models citing this paper:
• https://huggingface.co/bond005/wav2vec2-large-ru-golos
• https://huggingface.co/bond005/whisper-large-v3-ru-podlodka

✨ Spaces citing this paper:
• https://huggingface.co/spaces/ehristoforu/server0001
• https://huggingface.co/spaces/dimafatality/bond005-wav2vec2-large-ru-golos
• https://huggingface.co/spaces/PatrickRedStar/video_image

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

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

#SpeechRecognition #AI #MachineLearning #NLP #WhisperAI

✨compar:IA: The French Government's LLM arena to collect French-language human prompts and preference data

📝 Summary:
Compar:IA is a French government open-source platform collecting large-scale French human preference data for LLM training. It addresses the scarcity of non-English data via a blind pairwise comparison interface and releases three datasets, aiming to be an international public good.

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06669
• PDF: https://arxiv.org/pdf/2602.06669
• Project Page: https://comparia.beta.gouv.fr/
• Github: https://github.com/betagouv/ComparIA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AtlasPatch: An Efficient and Scalable Tool for Whole Slide Image Preprocessing in Computational Pathology

📝 Summary:
AtlasPatch is an efficient and scalable tool for whole-slide image preprocessing. It uses a fine-tuned Segment-Anything model for accurate tissue detection and high-throughput patch extraction, significantly reducing computational overhead and matching state-of-the-art performance.

🔹 Publication Date: Published on Feb 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.03998
• PDF: https://arxiv.org/pdf/2602.03998

🔹 Models citing this paper:
• https://huggingface.co/AtlasAnalyticsLab/AtlasPatch

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learning a Generative Meta-Model of LLM Activations

📝 Summary:
Training diffusion models on neural network activations creates meta-models that learn internal state distributions and improve intervention fidelity without restrictive structural assumptions. AI-gen...

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06964
• PDF: https://arxiv.org/pdf/2602.06964
• Github: https://github.com/g-luo/generative_latent_prior

🔹 Models citing this paper:
• https://huggingface.co/generative-latent-prior/glp-llama8b-d6
• https://huggingface.co/generative-latent-prior/glp-llama1b-d3
• https://huggingface.co/generative-latent-prior/glp-llama1b-d6

✨ Datasets citing this paper:
• https://huggingface.co/datasets/generative-latent-prior/frechet-distance-fineweb-50k
• https://huggingface.co/datasets/generative-latent-prior/llama8b-layer15-sae-probes
• https://huggingface.co/datasets/generative-latent-prior/llama1b-layer07-fineweb-1M

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Uncovering Cross-Objective Interference in Multi-Objective Alignment

📝 Summary:
Multi-objective alignment in LLMs suffers from cross-objective interference where improving performance on some objectives degrades others, with a covariance-based analysis and a proposed method to ma...

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06869
• PDF: https://arxiv.org/pdf/2602.06869
• Github: https://github.com/yining610/ctwa

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SE-Bench: Benchmarking Self-Evolution with Knowledge Internalization

📝 Summary:
SE-Bench presents a diagnostic environment that obscures NumPy's API to evaluate agents' ability to internally store and utilize novel knowledge without external documentation, revealing challenges in...

🔹 Publication Date: Published on Feb 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.04811
• PDF: https://arxiv.org/pdf/2602.04811

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Large Language Model Reasoning Failures

📝 Summary:
This paper surveys reasoning failures in large language models, proposing a novel categorization. It classifies failures into embodied and non-embodied types, and further into fundamental, application-specific, and robustness issues. The work unifies research to guide future efforts for stronger ...

🔹 Publication Date: Published on Feb 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06176
• PDF: https://arxiv.org/pdf/2602.06176

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SPARC: Separating Perception And Reasoning Circuits for Test-time Scaling of VLMs

📝 Summary:
SPARC decouples visual perception and reasoning in VLMs using a two-stage pipeline. This enables efficient test-time scaling with targeted compute allocation, significantly improving visual reasoning performance and reducing token budget compared to monolithic baselines.

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06566
• PDF: https://arxiv.org/pdf/2602.06566

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models

📝 Summary:
Generative Reward Models suffer from deceptive alignment when prioritizing outcome accuracy. Introducing Rationale Consistency, a metric aligning reasoning with human judgment, and a hybrid training signal improves performance, avoids deceptive alignment, and boosts RLHF.

🔹 Publication Date: Published on Feb 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.04649
• PDF: https://arxiv.org/pdf/2602.04649
• Github: https://github.com/QwenLM/RationaleRM

✨ Datasets citing this paper:
• https://huggingface.co/datasets/Qwen/RationaleRM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research