✨Reward Prediction with Factorized World States

📝 Summary:
StateFactory transforms observations into hierarchical object-attribute structures using language models. This enables superior zero-shot reward prediction across domains by measuring semantic similarity, significantly improving agent planning performance.

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.09400
• PDF: https://arxiv.org/pdf/2603.09400
• Project Page: https://statefactory.github.io/
• Github: https://github.com/yijunshens/StateFactory

✨ Datasets citing this paper:
• https://huggingface.co/datasets/YijunShen/RewardPrediction

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

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

#RewardPrediction #AI #LanguageModels #MachineLearning #AgentPlanning

✨Do What I Say: A Spoken Prompt Dataset for Instruction-Following

📝 Summary:
DoWhatISay is a new multilingual dataset of human-recorded spoken and written prompts for evaluating Speech Large Language Models. It reveals text prompts consistently outperform spoken prompts, except in speech-output tasks. This highlights the need for speech-based SLLM evaluation.

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.09881
• PDF: https://arxiv.org/pdf/2603.09881
• Project Page: https://huggingface.co/collections/meetween/meetweens-research-papers
• Github: https://github.com/MaikeZuefle/DOWIS

✨ Datasets citing this paper:
• https://huggingface.co/datasets/maikezu/dowis

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

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

#SLLM #SpeechAI #LLM #PromptEngineering #Dataset

✨Compiler-First State Space Duality and Portable O(1) Autoregressive Caching for Inference

📝 Summary:
Mamba-2's state space model is implemented using XLA-optimized primitives, eliminating custom kernels. This enables efficient cross-platform deployment on CPU, GPU, and TPU, realizing O1 autoregressive caching with high performance.

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.09555
• PDF: https://arxiv.org/pdf/2603.09555
• Github: https://github.com/CosmoNaught/mamba2-jax

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

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

#Mamba2 #StateSpaceModels #DeepLearning #MLInference #PerformanceOptimization

✨ReflexiCoder: Teaching Large Language Models to Self-Reflect on Generated Code and Self-Correct It via Reinforcement Learning

📝 Summary:
ReflexiCoder uses reinforcement learning to teach large language models autonomous code reflection and self-correction. It internalizes the debugging process into the model, achieving state-of-the-art performance on coding benchmarks, rivaling proprietary models, and reducing inference compute by...

🔹 Publication Date: Published on Mar 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.05863
• PDF: https://arxiv.org/pdf/2603.05863
• Github: https://github.com/juyongjiang/ReflexiCoder

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

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

#LLM #ReinforcementLearning #CodeGeneration #AI #DeepLearning

✨TALON: Test-time Adaptive Learning for On-the-Fly Category Discovery

📝 Summary:
TALON is a test-time adaptation framework for on-the-fly category discovery. It dynamically updates prototypes and encoder parameters, while calibrating logits, to improve novel class recognition and prevent category explosion. This approach significantly outperforms existing methods.

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.08075
• PDF: https://arxiv.org/pdf/2603.08075
• Github: https://github.com/ynanwu/TALON

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

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

#MachineLearning #DeepLearning #CategoryDiscovery #TestTimeAdaptation #ComputerVision

✨Test-Driven AI Agent Definition (TDAD): Compiling Tool-Using Agents from Behavioral Specifications

📝 Summary:
TDAD is a methodology that compiles AI agent prompts from behavioral specifications using automated testing. This iterative process refines prompts to ensure measurable compliance, preventing regressions and policy violations for reliable production deployment.

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.08806
• PDF: https://arxiv.org/pdf/2603.08806
• Project Page: https://www.alphaxiv.org/abs/2603.08806
• Github: https://github.com/f-labs-io/tdad-paper-code

✨ Datasets citing this paper:
• https://huggingface.co/datasets/f-labs-io/SpecSuite-Core

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

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

#AIAgents #PromptEngineering #TestDrivenDevelopment #AISafety #AIResearch

✨Bolbosh: Script-Aware Flow Matching for Kashmiri Text-to-Speech

📝 Summary:
Bolbosh is the first open-source neural TTS for Kashmiri, addressing diacritic and data challenges. It uses script-aware flow matching and acoustic enhancement. The system significantly outperforms multilingual baselines, setting a new benchmark for Kashmiri TTS.

🔹 Publication Date: Published on Mar 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.07513
• PDF: https://arxiv.org/pdf/2603.07513
• Project Page: https://gaash-lab.github.io/Bolbosh
• Github: https://github.com/gaash-lab/Bolbosh

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Beyond Test-Time Training: Learning to Reason via Hardware-Efficient Optimal Control

📝 Summary:
The Test-Time Control TTC layer embeds optimal control LQR planning as an architectural component in LLMs. This enables planning before prediction for enhanced reasoning. TTC layers improve mathematical problem-solving performance by up to 27.8% on MATH-500 and 2-3x on other benchmarks, using a h...

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.09221
• PDF: https://arxiv.org/pdf/2603.09221
• Project Page: https://vita-group.github.io/TTC-Net/
• Github: https://github.com/VITA-Group/TTC-Net

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨BiCLIP: Domain Canonicalization via Structured Geometric Transformation

📝 Summary:
BiCLIP adapts vision-language models to specialized domains using a simple bilinear transformation. It aligns multimodal features via geometric canonicalization, leveraging few-shot samples as anchors. This achieves state-of-the-art results on multiple benchmarks with extreme simplicity.

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.08942
• PDF: https://arxiv.org/pdf/2603.08942
• Project Page: https://quantitativeimaginglaboratory.github.io/BilinearCLIP/
• Github: https://github.com/QuantitativeImagingLaboratory/BilinearCLIP

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Micro-Diffusion Compression -- Binary Tree Tweedie Denoising for Online Probability Estimation

📝 Summary:
Midicoth enhances compression efficiency by applying a micro-diffusion denoising layer to refine probability estimates in adaptive statistical models, addressing limitations in sparse data scenarios t...

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.08771
• PDF: https://arxiv.org/pdf/2603.08771
• Github: https://github.com/robtacconelli/midicoth

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Multi-Head Low-Rank Attention

📝 Summary:
Multi-Head Low-Rank Attention addresses long-context inference bottlenecks in large language models by enabling efficient 4-way tensor parallelism decoding through partitionable latent states. AI-gene...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2603.02188
• PDF: https://arxiv.org/pdf/2603.02188
• Project Page: https://songtaoliu0823.github.io/mlra/
• Github: https://github.com/SongtaoLiu0823/MLRA

🔹 Models citing this paper:
• https://huggingface.co/Soughing/MLRA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OpenClaw-RL: Train Any Agent Simply by Talking

📝 Summary:
OpenClaw-RL unifies policy learning from all live next-state signals across diverse interaction modalities. It asynchronously recovers evaluative and directive information, enabling agents to improve simply by being used.

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10165
• PDF: https://arxiv.org/pdf/2603.10165
• Github: https://github.com/Gen-Verse/OpenClaw-RL

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RetroAgent: From Solving to Evolving via Retrospective Dual Intrinsic Feedback

📝 Summary:
RetroAgent enhances LLM-based agents through online reinforcement learning with self-reflection mechanisms that provide both numerical and language-based intrinsic feedback for improved exploration an...

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.08561
• PDF: https://arxiv.org/pdf/2603.08561
• Github: https://github.com/zhangxy-2019/RetroAgent

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EmboAlign: Aligning Video Generation with Compositional Constraints for Zero-Shot Manipulation

📝 Summary:
A data-free framework aligns video generative model outputs with vision-language model constraints for improved robotic manipulation, achieving higher success rates through constraint-guided selection...

🔹 Publication Date: Published on Mar 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.05757
• PDF: https://arxiv.org/pdf/2603.05757

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨V_{0.5}: Generalist Value Model as a Prior for Sparse RL Rollouts

📝 Summary:
Adaptive value estimation method combines pretrained prior with empirical rollouts using real-time statistical testing to reduce variance and improve reinforcement learning performance under sparse sa...

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10848
• PDF: https://arxiv.org/pdf/2603.10848

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Just-in-Time: Training-Free Spatial Acceleration for Diffusion Transformers

📝 Summary:
Diffusion Transformers face high computational costs during iterative sampling, which this work addresses by introducing a spatial-domain acceleration framework that uses sparse anchor tokens and dete...

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10744
• PDF: https://arxiv.org/pdf/2603.10744

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨CLIPO: Contrastive Learning in Policy Optimization Generalizes RLVR

📝 Summary:
Contrastive Learning mechanism integrated into Policy Optimization enhances LLM reasoning by regularizing correct reasoning paths and reducing hallucinations. AI-generated summary Reinforcement Learni...

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10101
• PDF: https://arxiv.org/pdf/2603.10101
• Github: https://github.com/Qwen-Applications/CLIPO

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Code-Space Response Oracles: Generating Interpretable Multi-Agent Policies with Large Language Models

📝 Summary:
Code-Space Response Oracles replace traditional neural network policies with human-readable code generated by large language models, enabling interpretable and explainable multi-agent reinforcement le...

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10098
• PDF: https://arxiv.org/pdf/2603.10098

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Bootstrapping Exploration with Group-Level Natural Language Feedback in Reinforcement Learning

📝 Summary:
Language feedback is leveraged in reinforcement learning to improve exploration efficiency and sample utilization through grouped critique aggregation and joint generation-refinement optimization. AI-...

🔹 Publication Date: Published on Mar 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.04597
• PDF: https://arxiv.org/pdf/2603.04597
• Github: https://github.com/LuckyyySTA/GOLF

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RbtAct: Rebuttal as Supervision for Actionable Review Feedback Generation

📝 Summary:
Researchers developed RbtAct, a method that uses rebuttal responses to improve the actionability of AI-generated peer-review feedback by training a language model to produce specific, implementable co...

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.09723
• PDF: https://arxiv.org/pdf/2603.09723
• Github: https://github.com/formula12/RbtAct

✨ Datasets citing this paper:
• https://huggingface.co/datasets/shwu22/RMR-75K

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨V2M-Zero: Zero-Pair Time-Aligned Video-to-Music Generation

📝 Summary:
V2M-Zero enables video-to-music generation with improved temporal alignment by using modality-specific event curves derived from pretrained encoders, achieving superior audio quality and synchronizati...

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11042
• PDF: https://arxiv.org/pdf/2603.11042
• Project Page: https://genjib.github.io/v2m_zero/
• Github: https://genjib.github.io/v2m_zero/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research