✨MegaFlow: Large-Scale Distributed Orchestration System for the Agentic Era

📝 Summary:
MegaFlow is a distributed orchestration system for large-scale AI agent training and evaluation. It addresses the lack of open-source infrastructure by providing efficient scheduling, resource allocation, and task management through modular services. MegaFlow successfully handles tens of thousand...

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07526
• PDF: https://arxiv.org/pdf/2601.07526

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Dr. Zero: Self-Evolving Search Agents without Training Data

📝 Summary:
A data-free self-evolution framework enables large language models to autonomously improve reasoning capabilities through iterative question generation and solving, achieving performance comparable to...

🔹 Publication Date: Published on Jan 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07055
• PDF: https://arxiv.org/pdf/2601.07055
• Github: https://github.com/facebookresearch/drzero

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts

📝 Summary:
Large reasoning models' inference latency can be reduced by routing reasoning steps to larger models based on the entropy of their first token, enabling efficient collaborative inference without addit...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05110
• PDF: https://arxiv.org/pdf/2601.05110
• Github: https://github.com/Zengwh02/GlimpRouter

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OpenTinker: Separating Concerns in Agentic Reinforcement Learning

📝 Summary:
OpenTinker provides a modular infrastructure for reinforcement learning of large language model agents with separated components and managed execution runtime. AI-generated summary We introduce OpenTi...

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2601.07376
• PDF: https://arxiv.org/pdf/2601.07376
• Project Page: https://open-tinker.github.io/opentinker-page/
• Github: https://github.com/open-tinker/OpenTinker

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨On the Fallacy of Global Token Perplexity in Spoken Language Model Evaluation

📝 Summary:
Speech models trained on raw audio can generate appropriate content while maintaining speaker and emotion attributes, but traditional text-based evaluation methods underestimate speech characteristics...

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.06329
• PDF: https://arxiv.org/pdf/2601.06329

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Are LLM Decisions Faithful to Verbal Confidence?

📝 Summary:
Large language models exhibit a disconnect between their expressed uncertainty and strategic decision-making under varying penalty conditions, failing to adjust abstention policies even when optimal. ...

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07767
• PDF: https://arxiv.org/pdf/2601.07767

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Codified Foreshadowing-Payoff Text Generation

📝 Summary:
Large language models struggle with maintaining long-range narrative dependencies, but a new framework called CFPG addresses this by structuring narrative continuity through executable causal predicat...

🔹 Publication Date: Published on Jan 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07033
• PDF: https://arxiv.org/pdf/2601.07033

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Controllable Memory Usage: Balancing Anchoring and Innovation in Long-Term Human-Agent Interaction

📝 Summary:
This paper presents SteeM, a framework for dynamically regulating memory reliance in LLM agents. It allows users to balance innovation with historical fidelity, overcoming the all-or-nothing problem of memory use. This approach outperforms conventional methods for personalized human-agent interac...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05107
• PDF: https://arxiv.org/pdf/2601.05107

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

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

#LLM #AI #HumanAgentInteraction #Memory #MachineLearning

✨DrivingGen: A Comprehensive Benchmark for Generative Video World Models in Autonomous Driving

📝 Summary:
DrivingGen is the first comprehensive benchmark for generative driving world models, addressing prior evaluation gaps. It uses diverse datasets and new metrics to assess visual realism, trajectory plausibility, temporal coherence, and controllability. Benchmarking reveals trade-offs between visua...

🔹 Publication Date: Published on Jan 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01528
• PDF: https://arxiv.org/pdf/2601.01528
• Project Page: https://drivinggen-bench.github.io/
• Github: https://github.com/youngzhou1999/DrivingGen

✨ Datasets citing this paper:
• https://huggingface.co/datasets/yangzhou99/DrivingGen

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

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

#AutonomousDriving #GenerativeAI #WorldModels #AIResearch #Benchmarking

✨"TODO: Fix the Mess Gemini Created": Towards Understanding GenAI-Induced Self-Admitted Technical Debt

📝 Summary:
Developers admit technical debt GIST in AI-assisted code, often due to postponed testing, incomplete adaptation, and limited understanding. This debt emerges when incorporating AI-generated code despite developer uncertainty about its behavior or correctness.

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07786
• PDF: https://arxiv.org/pdf/2601.07786

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OS-Symphony: A Holistic Framework for Robust and Generalist Computer-Using Agent

📝 Summary:
OS-Symphony is a framework enhancing computer-using agents with robustness and generalization. It features a Reflection-Memory Agent for self-correction and a Multimodal Searcher for visually aligned tutorials. This achieved state-of-the-art results on online benchmarks, including 65.84% on OSWorld.

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07779
• PDF: https://arxiv.org/pdf/2601.07779
• Project Page: https://os-copilot.github.io/OS-Symphony
• Github: https://github.com/OS-Copilot/OS-Symphony

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MHLA: Restoring Expressivity of Linear Attention via Token-Level Multi-Head

📝 Summary:
Multi-Head Linear Attention addresses the performance degradation in linear attention by preserving representational diversity through head-wise token dimension computation, maintaining linear complex...

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07832
• PDF: https://arxiv.org/pdf/2601.07832
• Project Page: https://dagroup-pku.github.io/MHLA/
• Github: https://github.com/DAGroup-PKU/MHLA

🔹 Models citing this paper:
• https://huggingface.co/DAGroup-PKU/MHLA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Beyond Hard Masks: Progressive Token Evolution for Diffusion Language Models

📝 Summary:
EvoToken-DLM introduces a diffusion-based language modeling approach that uses soft token distributions and continuous trajectory supervision to enable revisable decoding and outperforms existing base...

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07351
• PDF: https://arxiv.org/pdf/2601.07351
• Project Page: https://aim-uofa.github.io/EvoTokenDLM/
• Github: https://github.com/aim-uofa/EvoTokenDLM

🔹 Models citing this paper:
• https://huggingface.co/zhongzero/EvoToken_LLaDA_Instruct_8B_Lora

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨FinForge: Semi-Synthetic Financial Benchmark Generation

📝 Summary:
FinForge presents a scalable semi-synthetic pipeline for creating domain-specific financial evaluation benchmarks using expert curation and language model synthesis, demonstrating significant variatio...

🔹 Publication Date: Published on Jan 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.06747
• PDF: https://arxiv.org/pdf/2601.06747

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DrivingGen: A Comprehensive Benchmark for Generative Video World Models in Autonomous Driving

📝 Summary:
DrivingGen is the first comprehensive benchmark for generative driving world models, addressing prior evaluation gaps. It uses diverse datasets and new metrics to assess visual realism, trajectory plausibility, temporal coherence, and controllability. Benchmarking reveals trade-offs between visua...

🔹 Publication Date: Published on Jan 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01528
• PDF: https://arxiv.org/pdf/2601.01528
• Project Page: https://drivinggen-bench.github.io/
• Github: https://github.com/youngzhou1999/DrivingGen

✨ Datasets citing this paper:
• https://huggingface.co/datasets/yangzhou99/DrivingGen

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

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

#AutonomousDriving #GenerativeAI #WorldModels #AIResearch #Benchmarking

✨Gecko: An Efficient Neural Architecture Inherently Processing Sequences with Arbitrary Lengths

📝 Summary:
Gecko is a neural architecture for efficient processing of arbitrary length sequential data. It improves long range dependency capture with new components like timestep decay normalization and sliding chunk attention. Gecko outperforms Llama2 7B and Megalodon 7B, inherently handling sequences up ...

🔹 Publication Date: Published on Jan 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.06463
• PDF: https://arxiv.org/pdf/2601.06463
• Github: https://github.com/XuezheMax/gecko-llm

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

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

#Gecko #NeuralNetworks #SequenceModeling #LLM #DeepLearning

✨Forest Before Trees: Latent Superposition for Efficient Visual Reasoning

📝 Summary:
Laser introduces Dynamic Windowed Alignment Learning DWAL for visual reasoning. This method maintains global feature superposition, achieving state-of-the-art performance with significantly reduced computational costs and high efficiency.

🔹 Publication Date: Published on Jan 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.06803
• PDF: https://arxiv.org/pdf/2601.06803

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

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

#VisualReasoning #MachineLearning #AIResearch #ComputerVision #EfficientAI

✨FlyPose: Towards Robust Human Pose Estimation From Aerial Views

📝 Summary:
FlyPose is a lightweight, real-time aerial human pose estimation system. It achieves significantly improved accuracy through multi-dataset training and performs efficiently on UAVs. A new challenging dataset, FlyPose-104, is also released.

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05747
• PDF: https://arxiv.org/pdf/2601.05747
• Github: https://github.com/farooqhassaan/FlyPose

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

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

#HumanPoseEstimation #UAV #ComputerVision #DeepLearning #AI

✨mHC-lite: You Don't Need 20 Sinkhorn-Knopp Iterations

📝 Summary:
mHC-lite proposes a novel reparameterization for Hyper-Connections, explicitly constructing exactly doubly stochastic matrices via convex combinations of permutations. This approach guarantees stability, improves training throughput with native operations, and outperforms prior methods.

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05732
• PDF: https://arxiv.org/pdf/2601.05732
• Github: https://github.com/FFTYYY/mhc-lite

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

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

#DeepLearning #MachineLearning #Optimization #Algorithm #AI

✨Benchmarking Small Language Models and Small Reasoning Language Models on System Log Severity Classification

📝 Summary:
Severity classification benchmarks small language models for log understanding and deployability. RAG significantly boosts many models, even tiny ones, but efficiency and RAG integration vary widely, crucial for real-time systems.

🔹 Publication Date: Published on Jan 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.07790
• PDF: https://arxiv.org/pdf/2601.07790
• Github: https://github.com/stccenter/Benchmarking-SLMs-and-SRLMs-on-System-Log-Severity-Classification

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RealMem: Benchmarking LLMs in Real-World Memory-Driven Interaction

📝 Summary:
RealMem benchmark evaluates memory systems for long-term project-oriented interactions in large language models, revealing challenges in managing dynamic context dependencies. AI-generated summary As ...

🔹 Publication Date: Published on Jan 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.06966
• PDF: https://arxiv.org/pdf/2601.06966

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research