✨In-the-Flow Agentic System Optimization for Effective Planning and Tool Use

📝 Summary:
AgentFlow is a trainable agentic framework that optimizes its planner in-the-flow within multi-turn interactions. It uses Flow-GRPO to train its modules and significantly outperforms top baselines and GPT-4o on various reasoning and tool-use tasks.

🔹 Publication Date: Published on Oct 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2510.05592
• PDF: https://arxiv.org/pdf/2510.05592
• Project Page: https://agentflow.stanford.edu/
• Github: https://github.com/lupantech/AgentFlow

✨ Spaces citing this paper:
• https://huggingface.co/spaces/AgentFlow/agentflow
• https://huggingface.co/spaces/bioliveir4/agentflow2
• https://huggingface.co/spaces/bioliveir4/agentflow

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#AI #MachineLearning #AIagents #ToolUse #Planning

✨Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning

📝 Summary:
PaperCoder is a multi-agent LLM framework that automates converting machine learning papers into functional code repositories. It uses planning, analysis, and generation stages with specialized agents. Evaluations show it effectively creates high-quality implementations, outperforming strong base...

🔹 Publication Date: Published on Apr 24

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2504.17192
• PDF: https://arxiv.org/pdf/2504.17192
• Project Page: https://huggingface.co/papers/2504.15080
• Github: https://github.com/going-doer/Paper2Code

✨ Datasets citing this paper:
• https://huggingface.co/datasets/iaminju/paper2code

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#CodeGeneration #MachineLearning #LLM #AI #Automation

✨Grounded Misunderstandings in Asymmetric Dialogue: A Perspectivist Annotation Scheme for MapTask

📝 Summary:
This paper introduces a perspectivist annotation scheme for the MapTask corpus. It separately tracks speaker and addressee interpretations to reveal how understanding emerges and diverges. Findings show subtle discrepancies cause referential misalignment despite apparent agreement.

🔹 Publication Date: Published on Nov 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.03718
• PDF: https://arxiv.org/pdf/2511.03718

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#Dialogue #NLP #Communication #Pragmatics #CorpusLinguistics

✨DINOv3

📝 Summary:
DINOv3 is a self-supervised vision model excelling across tasks. It scales datasets, prevents dense feature degradation via Gram anchoring, and uses post-hoc strategies for flexibility. This versatile foundation model outperforms specialized state of the art without fine-tuning.

🔹 Publication Date: Published on Aug 13

🔹 Paper Links:
• arXiv Page: https://huggingface.co/collections/facebook/dinov3
• PDF: https://arxiv.org/pdf/2508.10104
• Project Page: https://ai.meta.com/blog/dinov3-self-supervised-vision-model/
• Github: https://github.com/facebookresearch/dinov3

🔹 Models citing this paper:
• https://huggingface.co/facebook/dinov3-vit7b16-pretrain-lvd1689m
• https://huggingface.co/facebook/dinov3-vitb16-pretrain-lvd1689m
• https://huggingface.co/facebook/dinov3-vitl16-pretrain-lvd1689m

✨ Datasets citing this paper:
• https://huggingface.co/datasets/zhuangzhe1229/test_dataset
• https://huggingface.co/datasets/simon123905/vitl

✨ Spaces citing this paper:
• https://huggingface.co/spaces/atalaydenknalbant/DINOv3
• https://huggingface.co/spaces/manu02/DINOv3-Interactive-Patch-Cosine-Similarity
• https://huggingface.co/spaces/merve/dinov3-viz

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#DINOv3 #SelfSupervisedLearning #ComputerVision #FoundationModels #AI

✨MarS: a Financial Market Simulation Engine Powered by Generative Foundation Model

📝 Summary:
MarS is a financial market simulation engine using LMM, an order-level generative model. It creates realistic, interactive market scenarios for risk-free strategy training and analysis. This offers scalability and strong realism.

🔹 Publication Date: Published on Sep 4, 2024

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2409.07486
• PDF: https://arxiv.org/pdf/2409.07486
• Github: https://github.com/microsoft/mars

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#FinancialMarkets #GenerativeAI #Simulation #LLM #FinTech

✨V-Thinker: Interactive Thinking with Images

📝 Summary:
V-Thinker is a multimodal reasoning assistant that enables interactive thinking with images using end-to-end reinforcement learning. It synthesizes datasets and aligns perception to enhance performance in vision-centric tasks, outperforming existing models.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04460
• PDF: https://arxiv.org/pdf/2511.04460
• Github: https://github.com/We-Math/V-Thinker

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#MultimodalAI #ComputerVision #ReinforcementLearning #AIResearch #DeepLearning

✨Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigm

📝 Summary:
The 'Thinking with Video' paradigm uses video generation models to unify multimodal reasoning, addressing limitations of static image or text-only approaches. Evaluated on VideoThinkBench, models like Sora-2 show strong performance on vision and text tasks, suggesting a promising unified reasonin...

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04570
• PDF: https://arxiv.org/pdf/2511.04570

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#VideoGeneration #MultimodalAI #AIResearch #ComputerVision #DeepLearning

✨GUI-360: A Comprehensive Dataset and Benchmark for Computer-Using Agents

📝 Summary:
GUI-360 is a large dataset and benchmark for computer-using agents, addressing gaps in real-world tasks and unified evaluation. It contains over 1.2M action steps in Windows apps for GUI grounding, screen parsing, and action prediction. Benchmarking reveals significant shortcomings in current mod...

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04307
• PDF: https://arxiv.org/pdf/2511.04307

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#AI #ComputerAgents #GUIAgents #Dataset #Benchmark

✨Cambrian-S: Towards Spatial Supersensing in Video

📝 Summary:
This paper promotes spatial supersensing for AI, including predictive world modeling. It introduces VSI-SUPER and a predictive sensing method leveraging surprise for memory, outperforms baselines, showing anticipation is vital.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04670
• PDF: https://arxiv.org/pdf/2511.04670
• Project Page: https://cambrian-mllm.github.io/

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#AI #ComputerVision #PredictiveModeling #MachineLearning #SpatialSensing

✨The Strong Lottery Ticket Hypothesis for Multi-Head Attention Mechanisms

📝 Summary:
This paper theoretically proves the strong lottery ticket hypothesis for multi-head attention mechanisms, showing SLTs exist with sufficient hidden dimensions. It extends the hypothesis to transformers without normalization layers, with empirical validation.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04217
• PDF: https://arxiv.org/pdf/2511.04217

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#LotteryTicketHypothesis #MultiHeadAttention #Transformers #DeepLearning #NeuralNetworks

✨NVIDIA Nemotron Nano V2 VL

📝 Summary:
Nemotron Nano V2 VL is a new hybrid Mamba-Transformer LLM designed for improved document and video understanding. It leverages enhanced architecture and token reduction for higher inference throughput.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.03929
• PDF: https://arxiv.org/pdf/2511.03929

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#LLM #MambaTransformer #MultimodalAI #AIResearch #DeepLearning

✨Scaling Agent Learning via Experience Synthesis

📝 Summary:
DreamGym is a unified framework that synthesizes diverse experiences for scalable online reinforcement learning. It distills environment dynamics into a reasoning-based model to reduce reliance on expensive real-world rollouts. DreamGym significantly improves RL training performance and reduces t...

🔹 Publication Date: Published on Nov 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.03773
• PDF: https://arxiv.org/pdf/2511.03773

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#ReinforcementLearning #MachineLearning #AI #AgentLearning #ExperienceSynthesis

✨Benchmark Designers Should "Train on the Test Set" to Expose Exploitable Non-Visual Shortcuts

📝 Summary:
Multimodal benchmarks are vulnerable to models exploiting non-visual shortcuts. This paper proposes designers train on the test set to diagnose and mitigate these biases, leading to more robust benchmarks for MLLM evaluation and revealing widespread issues.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04655
• PDF: https://arxiv.org/pdf/2511.04655
• Project Page: https://cambrian-mllm.github.io/

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#MultimodalAI #BenchmarkDesign #AIbias #MLLMEvaluation #RobustAI

✨Learning Vision-Driven Reactive Soccer Skills for Humanoid Robots

📝 Summary:
A unified reinforcement learning controller directly integrates visual perception and motion control for humanoid soccer robots. It uses extended Adversarial Motion Priors and an encoder-decoder to achieve reactive, coherent, and robust soccer skills in dynamic real-world environments.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.03996
• PDF: https://arxiv.org/pdf/2511.03996
• Project Page: https://humanoid-kick.github.io/

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#HumanoidRobots #ReinforcementLearning #Robotics #ComputerVision #AI

✨Contamination Detection for VLMs using Multi-Modal Semantic Perturbation

📝 Summary:
This paper introduces a novel method to detect contamination in Vision-Language Models. It uses multi-modal semantic perturbation, showing that contaminated models fail to generalize under controlled changes. The method is robust across diverse contamination strategies.

🔹 Publication Date: Published on Nov 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.03774
• PDF: https://arxiv.org/pdf/2511.03774

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#VLM #AIContamination #MultiModalAI #MachineLearning #AIResearch

✨How to Evaluate Speech Translation with Source-Aware Neural MT Metrics

📝 Summary:
This study introduces source-aware metrics for speech translation evaluation by generating text proxies from audio, like ASR trannoscripts or back-translations. A new re-segmentation algorithm resolves alignment issues. These methods improve evaluation accuracy for speech translation systems.

🔹 Publication Date: Published on Nov 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.03295
• PDF: https://arxiv.org/pdf/2511.03295

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#SpeechTranslation #NMTMetrics #ASR #NLP #DeepLearning

✨RDMA Point-to-Point Communication for LLM Systems

📝 Summary:
TransferEngine provides a uniform interface for flexible point-to-point communication in LLM systems, overcoming NIC-specific limitations. It bridges different hardware, providing high throughput for disaggregated inference, RL, and MoE tasks. This solution avoids hardware lock-in and complements...

🔹 Publication Date: Published on Oct 31

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2510.27656
• PDF: https://arxiv.org/pdf/2510.27656
• Github: https://github.com/perplexityai/pplx-garden

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#RDMA #LLM #HPC #AIInfrastructure #DistributedSystems

✨SAIL-RL: Guiding MLLMs in When and How to Think via Dual-Reward RL Tuning

📝 Summary:
SAIL-RL uses a dual-reward RL system to teach MLLMs when and how to think. This improves reasoning, reduces hallucinations, and achieves competitive performance against commercial models like GPT-4o.

🔹 Publication Date: Published on Nov 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.02280
• PDF: https://arxiv.org/pdf/2511.02280

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#MLLMs #ReinforcementLearning #AI #GenerativeAI #DeepLearning

✨SIMS-V: Simulated Instruction-Tuning for Spatial Video Understanding

📝 Summary:
SIMS-V uses 3D simulators to generate diverse spatial video training data. This efficiently trains multimodal language models, overcoming real-world data bottlenecks. A 7B model trained on this simulated data significantly outperforms larger baselines on real-world spatial reasoning tasks.

🔹 Publication Date: Published on Nov 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.04668
• PDF: https://arxiv.org/pdf/2511.04668
• Github: https://ellisbrown.github.io/sims-v/

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#SpatialAI #MultimodalLLM #SimulatedData #ComputerVision #DeepLearning

✨EVTAR: End-to-End Try on with Additional Unpaired Visual Reference

📝 Summary:
EVTAR is an end-to-end virtual try-on model that enhances accuracy and garment detail preservation using additional reference images. It simplifies the process by requiring only source and target garment inputs, producing high-quality, realistic try-on results.

🔹 Publication Date: Published on Nov 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.00956
• PDF: https://arxiv.org/pdf/2511.00956
• Github: https://github.com/360CVGroup/EVTAR

🔹 Models citing this paper:
• https://huggingface.co/qihoo360/EVTAR

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#VirtualTryOn #ComputerVision #DeepLearning #AIFashion #ImageSynthesis