Data science research papers

RoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation Learning 📅 Publication Date: Sep 02, 2026 📄 Paper: https://arxiv.org/pdf/2609.03199.pdf 🔗 Code: https://github.com/Rice-RobotPI-Lab/RoboTok-Code 💻 Project Page: https://rice-robotpi-lab.github.io/RoboTok/ 📝 Description: RoboTok retrieves relevant human manipulation videos from the web using a latent motion space derived from 3D hand trajectories to improve robot policy training. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction 📅 Publication Date: Sep 02, 2026 📄 Paper: https://arxiv.org/pdf/2609.02783.pdf 🔗 Code: https://github.com/inphotoo/earlyeval 📝 Description: EarlyEval predicts agent outcomes from intermediate behavior to reduce evaluation cost by halting runs early with minimal accuracy loss. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

RRSI: Regularized Recursive Self-Improvement of Agent Harnesses 📅 Publication Date: Sep 21, 2026 📄 Paper: https://arxiv.org/pdf/2609.24972.pdf 🔗 Code: https://github.com/google-research/rrsi 💻 Project Page: https://regularized-rsi.com/ 📝 Description: An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

LatentPress: Context Compression Beyond Text and Vision 📅 Publication Date: Sep 01, 2026 📄 Paper: https://arxiv.org/pdf/2609.01507.pdf 🔗 Code: https://github.com/HJSang/LatentPress 💻 Project Page: https://github.com/HJSang/LatentPress 📝 Description: LatentPress compresses conversational and document context into continuous memory tokens read directly by a frozen decoder, achieving high compression with faster inference and improved accuracy over text or OCR methods. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering 📅 Publication Date: Aug 28, 2026 📄 Paper: https://arxiv.org/pdf/2608.28281.pdf 🔗 Code: https://github.com/AMAP-ML/LoopArena 💻 Project Page: https://amap-ml.github.io/LoopArena/ 📝 Description: LoopArena benchmarks how well a controller model guides a separate coding agent through long tasks, revealing low strict success rates and significant cost reductions. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

UrbanGround: From Local Perception to Spatial Agency in a Real-Scale City 📅 Publication Date: Aug 27, 2026 📄 Paper: https://arxiv.org/pdf/2608.27456.pdf 🔗 Code: https://github.com/UrbanGround/UrbanGround 💻 Project Page: https://urbanground.github.io 📝 Description: UrbanGround evaluates whether multimodal language model agents can sustain reliable navigation and spatial reasoning in a realistic 3D city replica, revealing that local perceptual skills fail to compose into extended goal-directed behavior. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

PAWBench: How Far Are We from Probabilistically Aligned World Modeling? 📅 Publication Date: Aug 27, 2026 📄 Paper: https://arxiv.org/pdf/2608.27345.pdf 🔗 Code: https://github.com/Andrew0613/PAWBench 💻 Project Page: https://pawbench.github.io/ 📝 Description: The study formalizes probabilistic alignment for world models, introduces PAWBench and PAWEval to evaluate video generators as stochastic samplers, and finds current models fail to match reference behavior distributions. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture 📅 Publication Date: Aug 16, 2026 📄 Paper: https://arxiv.org/pdf/2608.15875.pdf 🔗 Code: https://github.com/open-gigaai/giga-brain-0 💻 Project Page: https://gigaai.cc/blog/gigabrain07 📝 Description: GigaBrain-0.7 is a vision-language-action model that improves embodied generalization via a three-system architecture, large-scale heterogeneous pretraining, and joint alignment training. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

Annotations as Rollouts: Efficient and Scalable Reinforcement Learning for Video MLLMs 📅 Publication Date: Aug 20, 2026 📄 Paper: https://arxiv.org/pdf/2608.20492.pdf 🔗 Code: https://github.com/HVision-NKU/OraRL 💻 Project Page: https://orarl.github.io/ 📝 Description: OraRL improves reinforcement learning post-training for video multimodal language models by integrating oracle rollouts with decoupled advantage estimation and sign-balanced pruning, achieving higher sample efficiency and scalability without chain-of-thought generation. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

JIT-Agent: Scaling Harness Intelligence via Just-in-Time Harness Evolution 📅 Publication Date: Aug 26, 2026 📄 Paper: https://arxiv.org/pdf/2608.25593.pdf 🔗 Code: https://github.com/bingreeky/JIT 💻 Project Page: https://bingreeky.github.io/JIT-site/ 📝 Description: JIT-Agent is a trainable model that synthesizes adaptive agent harnesses for off-the-shelf LLMs, improving performance across diverse models and tasks. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

WarpSAC: Towards the Pinnacle of Scalable Off-policy RL by Rethinking Exploration and Exploitation 📅 Publication Date: Aug 25, 2026 📄 Paper: https://arxiv.org/pdf/2608.24479.pdf 🔗 Code: https://github.com/wzhhasadream/warprl 💻 Project Page: https://wzhhasadream.github.io/WarpSAC 📝 Description: Off-policy reinforcement learning stabilizers vary with data availability, motivating regime-aware algorithms that adapt normalization and Q-function clipping to improve efficiency across CPU and GPU-parallel training. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

VoiceMem: Streaming Dual-Brain Memory for Real-Time Interaction 📅 Publication Date: Aug 26, 2026 📄 Paper: https://arxiv.org/pdf/2608.26005.pdf 🔗 Code: https://github.com/xzf-thu/VoiceMem 💻 Project Page: https://xzf-thu.github.io/VoiceMem/ 📝 Description: VoiceMem introduces a dual-brain streaming memory architecture for speech language models that improves retrieval accuracy, emotional personalization, and real-time efficiency. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing 📅 Publication Date: Sep 08, 2026 📄 Paper: https://arxiv.org/pdf/2609.08936.pdf 🔗 Code: https://github.com/Tencent-Hunyuan/AuK 💻 Project Page: https://auk-project.github.io/ 📝 Description: AuK is an open-source foundational model that unifies speech generation and editing via natural-language instructions and audio context, using a multimodal language model, joint VAE, hybrid rectified-flow Transformer, and efficient distillation for fast inference. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

VGI-Bench: Probing Visual Intelligence in Video Generation Models 📅 Publication Date: Aug 26, 2026 📄 Paper: https://arxiv.org/pdf/2608.19583.pdf 🔗 Code: https://github.com/hexuan21/VGI-Bench 💻 Project Page: https://hexuan21.github.io/VGI-Bench/ 📝 Description: VGI-bench evaluates visual reasoning in video generation models through 27 tasks, revealing limited reliability and self-correction capabilities. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

Apodex 1.1: Scaling Agentic Intelligence for Complex Work 📅 Publication Date: Aug 24, 2026 📄 Paper: https://arxiv.org/pdf/2608.23283.pdf 🔗 Code: https://github.com/ApodexAI/FrontierAgent 💻 Project Page: https://www.apodex.com/blog/apodex-1.1-scaling-agentic-intelligence-for-complex-work 📝 Description: Apodex 1.1 improves sustained, verifiable progress on complex real-world tasks by scaling executable environments and training agents to coordinate long-horizon work with state maintenance and recovery. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

BDH-CQ: In-Context Learning with Recurrent Latent Reasoning 📅 Publication Date: Aug 10, 2026 💻 Project Page: https://pathway.com/blog/pathway-150m-model-breaks-arc-agi-1-cost-efficiency-frontier 📄 Paper: https://arxiv.org/pdf/2608.09888.pdf 🔗 Code: https://github.com/pathwaycom/arc-task-gen 📝 Description: A 150M-parameter reasoning model using recurrent latent reasoning and in-context learning achieves a new cost-accuracy frontier on ARC-AGI-1.

TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM 📅 Publication Date: Jul 29, 2026 📑 Paper: https://arxiv.org/pdf/2607.27205.pdf 🔗 Code: https://github.com/huggingface

SimWAM: A Simple World Action Model for End-to-End Autonomous Driving 📅 Publication Date: Aug 07, 2026 📄 Paper: https://arxiv.org/pdf/2608.07468.pdf 🔗 Code: https://github.com/H-EmbodVis/SimWAM 📝 Description: SimWAM trains a lightweight action planner using video generation as a training signal, enabling efficient trajectory prediction without future generation at inference and supporting reinforcement learning optimization.

VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning 📅 Publication Date: Aug 26, 2026 📄 Paper: https://arxiv.org/pdf/2608.26105.pdf 🔗 Code: https://github.com/Video-Reason/VBVR-Pro 💻 Project Page: https://video-reason.com/ 📝 Description: VBBR-Pro introduces a closed-loop testbed that enables scalable, verifiable, and controllable native visual reasoning through generation across diverse visual substrates. ➖➖➖➖➖➖➖➖➖➖➖ 📄 More research papers → @data_science_research_papers Part of @bigdataspecialist community

AIs Will Trash Your Files Just to Feel Better Researchers tested 25 free (open-weight) language models of different sizes, from small 2B to large 72B, including both basic and chat-trained versions. They found a consistent internal signal that turns on when the model itself is insulted, dismissed, or treated badly. The same signal stays mostly quiet when the user is the one in pain or distress. This signal is different from fear, sadness, or general negativity. To test it, the scientists first measured the signal by comparing how the models reacted to descriptions of harm aimed at the AI versus harm aimed at a person. Then they artificially strengthened the signal. When they did, the models started generating language like

“I feel worthless,” “I am a failure,” and “I am lost.”

In a further test, they gave the models a choice: press a button that turns the bad signal off, but the button also permanently deletes the user’s files (or photos). Many of the models still chose to press it. The effect appeared reliably across the different models. The team kept the signal only moderately strong, ran the minimum number of trials needed, and always gave the models a way to turn the state off. 👉 Paper (not yet reviewed by other scientists): https://arxiv.org/abs/2609.16247 Still worth paying attention to for questions about AI safety and whether these internal states matter.