✨Toward Autonomous Long-Horizon Engineering for ML Research

📝 Summary:
AiScientist enables autonomous long-horizon ML research engineering by combining hierarchical orchestration with durable state management, achieving superior performance on benchmark tasks through str...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13018
• PDF: https://arxiv.org/pdf/2604.13018
• Github: https://github.com/AweAI-Team/AiScientist

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks

📝 Summary:
Sequence-Level PPO addresses instability in long-chain-of-thought reasoning by reformulating the process as a contextual bandit problem with decoupled value functions for improved efficiency. AI-gener...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.08865
• PDF: https://arxiv.org/pdf/2604.08865
• Github: https://github.com/sustech-nlp/SPPO

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Habitat-GS: A High-Fidelity Navigation Simulator with Dynamic Gaussian Splatting

📝 Summary:
Habitat-GS extends Habitat-Sim by integrating 3D Gaussian Splatting for photorealistic rendering and gaussian avatars for dynamic human modeling, enabling improved agent generalization and human-aware...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12626
• PDF: https://arxiv.org/pdf/2604.12626
• Project Page: https://zju3dv.github.io/habitat-gs/
• Github: https://github.com/zju3dv/habitat-gs

✨ Datasets citing this paper:
• https://huggingface.co/datasets/RukawaY/gs_scenes

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Rethinking the Diffusion Model from a Langevin Perspective

📝 Summary:
The article provides a unified Langevin perspective on diffusion models, clarifying their theoretical foundations and connections between different mathematical formulations. AI-generated summary Diff...

🔹 Publication Date: Published on Apr 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10465
• PDF: https://arxiv.org/pdf/2604.10465
• Project Page: https://iclr-blogposts.github.io/2026/blog/2026/rethinking-diffusion-langevin/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Generative Refinement Networks for Visual Synthesis

📝 Summary:
Generative Refinement Networks introduce a novel visual synthesis approach that combines hierarchical binary quantization with adaptive refinement mechanisms to improve computational efficiency and vi...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13030
• PDF: https://arxiv.org/pdf/2604.13030
• Github: https://github.com/MGenAI/GRN

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation

📝 Summary:
Lightning OPD enables efficient offline on-policy distillation for large language models by enforcing teacher consistency and eliminating the need for live teacher inference servers. AI-generated summ...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13010
• PDF: https://arxiv.org/pdf/2604.13010

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

📝 Summary:
Nemotron 3 Super is a 120 billion parameter hybrid Mamba-Attention Mixture-of-Experts model pre-trained in NVFP4 with LatentMoE architecture and MTP layers for accelerated inference, achieving superio...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12374
• PDF: https://arxiv.org/pdf/2604.12374

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Towards Long-horizon Agentic Multimodal Search

📝 Summary:
A novel long-horizon multimodal deep search framework called LMM-Searcher is introduced, featuring a file-based visual representation mechanism and progressive visual loading to handle heterogeneous i...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12890
• PDF: https://arxiv.org/pdf/2604.12890
• Github: https://github.com/RUCAIBox/LMM-Searcher

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨KnowRL: Boosting LLM Reasoning via Reinforcement Learning with Minimal-Sufficient Knowledge Guidance

📝 Summary:
KnowRL is a knowledge-guided reinforcement learning framework that improves reasoning in language models by optimizing compact, interaction-aware guidance subsets through constrained subset search and...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12627
• PDF: https://arxiv.org/pdf/2604.12627
• Github: https://github.com/Hasuer/KnowRL

🔹 Models citing this paper:
• https://huggingface.co/HasuerYu/KnowRL-Nemotron-1.5B

✨ Datasets citing this paper:
• https://huggingface.co/datasets/HasuerYu/KnowRL-KP-Annotations
• https://huggingface.co/datasets/HasuerYu/KnowRL-Train-Data

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Self-Adversarial One Step Generation via Condition Shifting

📝 Summary:
APEx enables efficient one-step text-to-image synthesis by eliminating adversarial training through endogenous gradient estimation from flow models, achieving superior quality and speed compared to ex...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12322
• PDF: https://arxiv.org/pdf/2604.12322
• Github: https://github.com/LINs-lab/APEX

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Lyra 2.0: Explorable Generative 3D Worlds

📝 Summary:
Lyra 2.0 enables large-scale 3D scene creation through persistent video generation that addresses spatial forgetting and temporal drifting issues in long-horizon video models. AI-generated summary Rec...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13036
• PDF: https://arxiv.org/pdf/2604.13036

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨LARY: A Latent Action Representation Yielding Benchmark for Generalizable Vision-to-Action Alignment

📝 Summary:
General visual foundation models trained without action supervision outperform specialized embodied models and demonstrate superior alignment between visual and physical action spaces compared to pixe...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11689
• PDF: https://github.com/meituan-longcat/LARYBench/blob/main/LARYBench.pdf
• Project Page: https://meituan-longcat.github.io/LARYBench/
• Github: https://meituan-longcat.github.io/LARYBench/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨You Only Judge Once: Multi-response Reward Modeling in a Single Forward Pass

📝 Summary:
A multimodal reward model evaluates multiple responses simultaneously through concatenated input and cross-entropy scoring, achieving faster training and superior performance in open-ended generation ...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10966
• PDF: https://arxiv.org/pdf/2604.10966

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨VideoFlexTok: Flexible-Length Coarse-to-Fine Video Tokenization

📝 Summary:
VideoFlexTok enables efficient video representation through variable-length token sequences that capture abstract information first, followed by fine-grained details, allowing for reduced computationa...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12887
• PDF: https://arxiv.org/pdf/2604.12887
• Github: https://github.com/apple/ml-videoflextok

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Seeing Through Touch: Tactile-Driven Visual Localization of Material Regions

📝 Summary:
Deep learning model for tactile localization that uses dense cross-modal feature interactions to identify material properties in images, overcoming limitations of existing methods through enhanced dat...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11579
• PDF: https://arxiv.org/pdf/2604.11579
• Project Page: https://mm.kaist.ac.kr/projects/SeeingThroughTouch/
• Github: https://github.com/kaistmm/SeeingThroughTouch

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Many-Tier Instruction Hierarchy in LLM Agents

📝 Summary:
Large language model agents require robust instruction conflict resolution mechanisms that can handle arbitrary privilege levels across diverse real-world scenarios, revealing current models' limitati...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09443
• PDF: https://arxiv.org/pdf/2604.09443
• Project Page: https://jhu-clsp.github.io/ManyIH
• Github: https://github.com/JHU-CLSP/ManyIH

✨ Datasets citing this paper:
• https://huggingface.co/datasets/jhu-clsp/ManyIH-Bench
• https://huggingface.co/datasets/jackzhang/ManyIH-Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Hierarchical SVG Tokenization: Learning Compact Visual Programs for Scalable Vector Graphics Modeling

📝 Summary:
HiVG introduces a hierarchical SVG tokenization framework that improves autoregressive vector graphics generation by addressing geometric structure representation and spatial consistency issues throug...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.05072
• PDF: https://arxiv.org/pdf/2604.05072
• Project Page: https://hy-hivg.github.io/
• Github: https://github.com/ximinng/HiVG

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Beyond Perception Errors: Semantic Fixation in Large Vision-Language Models

📝 Summary:
Vision-language models exhibit semantic fixation by preferring default interpretations over alternative valid rule mappings, which can be mitigated through prompt interventions and training strategies...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12119
• PDF: https://arxiv.org/pdf/2604.12119
• Project Page: https://maveryn.github.io/vlm-fix/
• Github: https://github.com/maveryn/vlm-fix

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨CONSCIENTIA: Can LLM Agents Learn to Strategize? Emergent Deception and Trust in a Multi-Agent NYC Simulation

📝 Summary:
Large language model agents demonstrate limited strategic behaviors including selective trust and deception in a simulated urban environment, remaining vulnerable to adversarial persuasion despite imp...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09746
• PDF: https://arxiv.org/pdf/2604.09746

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Masked by Consensus: Disentangling Privileged Knowledge in LLM Correctness

📝 Summary:
LLMs generally lack superior self-awareness of correctness. However, when models disagree, they demonstrate privileged knowledge for factual tasks, outperforming peers. This advantage emerges in early-to-mid layers, but not in math reasoning.

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12373
• PDF: https://arxiv.org/pdf/2604.12373
• Project Page: https://technion-cs-nlp.github.io/Privileged-Knowledge/

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

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

#LLM #AI #NLP #ModelCorrectness #PrivilegedKnowledge

✨Accelerating Speculative Decoding with Block Diffusion Draft Trees

📝 Summary:
DDTree enhances speculative decoding by constructing draft trees from block diffusion drafter distributions. It efficiently verifies multiple trajectories in parallel in a single target model pass, improving performance.

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12989
• PDF: https://arxiv.org/pdf/2604.12989
• Project Page: https://liranringel.github.io/ddtree
• Github: https://github.com/liranringel/ddtree

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

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

#SpeculativeDecoding #BlockDiffusion #LLMAcceleration #DeepLearning #AIResearch