ML Research Hub
32.3K subscribers
6.73K photos
472 videos
24 files
7.34K links
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.

Admin: @HusseinSheikho || @Hussein_Sheikho
Download Telegram
SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation

📝 Summary:
A modular framework called SPASM is presented for generating stable multi-turn dialogues with consistent personas, addressing issues like persona drift and echoing through a perspective-agnostic conte...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09212
• PDF: https://arxiv.org/pdf/2604.09212
• Project Page: https://arxiv.org/abs/2604.09212
• Github: https://github.com/lhannnn/SPASM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
SWE-AGILE: A Software Agent Framework for Efficiently Managing Dynamic Reasoning Context

📝 Summary:
SWE-AGILE addresses reasoning limitations in software engineering by using dynamic context management to balance detailed analysis with computational efficiency. AI-generated summary Prior representat...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11716
• PDF: https://arxiv.org/pdf/2604.11716
• Github: https://github.com/KDEGroup/SWE-AGILE

🔹 Models citing this paper:
https://huggingface.co/KDEGroup/SWE-AGILE-RL-8B

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Uni-ViGU: Towards Unified Video Generation and Understanding via A Diffusion-Based Video Generator

📝 Summary:
Uni-ViGU introduces a unified framework for video generation and understanding, uniquely building upon a video generator as its foundation. It uses unified flow matching and a bidirectional training mechanism to achieve competitive performance in both generation and understanding tasks.

🔹 Publication Date: Published on Apr 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.08121
• PDF: https://arxiv.org/pdf/2604.08121
• Project Page: https://fr0zencrane.github.io/uni-vigu-page/
• Github: https://fr0zencrane.github.io/uni-vigu-page/

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

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

#VideoGeneration #VideoUnderstanding #DiffusionModels #AIResearch #DeepLearning
DiningBench: A Hierarchical Multi-view Benchmark for Perception and Reasoning in the Dietary Domain

📝 Summary:
A new hierarchical, multi-view benchmark called DiningBench is introduced to evaluate vision-language models on fine-grained food classification, nutrition estimation, and visual question answering, r...

🔹 Publication Date: Published on Apr 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10425
• PDF: https://arxiv.org/pdf/2604.10425
• Github: https://github.com/meituan/DiningBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Playing Along: Learning a Double-Agent Defender for Belief Steering via Theory of Mind

📝 Summary:
Large language models face challenges in theory-of-mind reasoning for adversarial interactions, but reinforcement learning-trained AI double agents demonstrate improved belief manipulation and theory-...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11666
• PDF: https://arxiv.org/pdf/2604.11666
• Github: https://github.com/The-Inscrutable-X/AIDoubleAgentDefenders

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding

📝 Summary:
SPEED-Bench is introduced as a new benchmark for Speculative Decoding SD evaluation. It provides diverse semantic domains and realistic serving regimes to address limitations of existing benchmarks. This enables accurate measurement of SD performance in production environments, setting a unified ...

🔹 Publication Date: Published on Feb 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09557
• PDF: https://arxiv.org/pdf/2604.09557
• Project Page: https://huggingface.co/blog/nvidia/speed-bench
• Github: https://github.com/NVIDIA/Model-Optimizer

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

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

#SpeculativeDecoding #AIBenchmarks #LLMs #DeepLearning #ModelOptimization
Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models

📝 Summary:
Masked diffusion language models can be accelerated through strategic replacement of full models with smaller ones during specific denoising steps, achieving reduced computational costs with minimal i...

🔹 Publication Date: Published on Apr 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02340
• PDF: https://arxiv.org/pdf/2604.02340

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Polyglot Teachers: Evaluating Language Models for Multilingual Synthetic Data Generation

📝 Summary:
Effective multilingual teacher models for synthetic data generation are identified through systematic evaluation of data quality metrics rather than model size alone, with findings showing that prompt...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11290
• PDF: https://arxiv.org/pdf/2604.11290
• Github: https://github.com/ljvmiranda921/polyglot-teachers

🔹 Models citing this paper:
https://huggingface.co/ljvmiranda921/Polyglot-Gemma3-4B-SFT-ar
https://huggingface.co/ljvmiranda921/Polyglot-OLMo3-7B-SFT-ar
https://huggingface.co/ljvmiranda921/Polyglot-OLMo3-7B-SFT-cs

Datasets citing this paper:
https://huggingface.co/datasets/ljvmiranda921/PolyglotTeachers-SFT-Synth

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Enhancing Financial Sentiment Analysis via Retrieval Augmented Large Language Models

📝 Summary:
A retrieval-augmented LLM framework improves financial sentiment analysis by tuning LLMs for sentiment prediction and augmenting them with external context, outperforming traditional models and other ...

🔹 Publication Date: Published on Oct 6, 2023

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2310.04027
• PDF: https://arxiv.org/pdf/2310.04027
• Github: https://github.com/AI4Finance-Foundation/FinGPT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
QuanBench+: A Unified Multi-Framework Benchmark for LLM-Based Quantum Code Generation

📝 Summary:
QuanBench+ evaluates large language models on quantum code generation across multiple frameworks using functional testing and repair-based feedback, revealing significant progress but persistent depen...

🔹 Publication Date: Published on Mar 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.08570
• PDF: https://arxiv.org/pdf/2604.08570
• Github: https://github.com/JawadKotaichh/quanbench-plus

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
BMdataset: A Musicologically Curated LilyPond Dataset

📝 Summary:
A curated LilyPond dataset and adapted CodeBERT model demonstrate that expert-curated small datasets can outperform large noisy corpora for music understanding tasks. AI-generated summary Symbolic mus...

🔹 Publication Date: Published on Apr 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10628
• PDF: https://arxiv.org/pdf/2604.10628
• Project Page: https://zenodo.org/records/18723290
• Github: https://github.com/CSCPadova/lilybert

🔹 Models citing this paper:
https://huggingface.co/csc-unipd/lilybert

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Advancing Polish Language Modeling through Tokenizer Optimization in the Bielik v3 7B and 11B Series

📝 Summary:
The Bielik v3 PL series achieves improved language-specific performance through specialized Polish tokenization, FOCUS-based embeddings, and multi-stage training with supervised fine-tuning, direct pr...

🔹 Publication Date: Published on Apr 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10799
• PDF: https://arxiv.org/pdf/2604.10799
• Project Page: https://bielik.ai/

🔹 Models citing this paper:
https://huggingface.co/speakleash/Bielik-PL-11B-v3.0-Instruct
https://huggingface.co/speakleash/Bielik-PL-Minitron-7B-v3.0-Instruct

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping

📝 Summary:
MEDS improves RL for LLMs by addressing reduced sampling diversity. It uses historical behavioral signals and clustering to identify and penalize recurrent error patterns, encouraging broader exploration. This framework consistently boosts performance and behavioral diversity during sampling.

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11297
• PDF: https://arxiv.org/pdf/2604.11297
• Github: https://github.com/Linxi000/MEDS

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Mobile GUI Agent Privacy Personalization with Trajectory Induced Preference Optimization

📝 Summary:
Mobile GUI agents neglect user privacy personalization, as varied execution trajectories hinder standard optimization. This paper proposes Trajectory Induced Preference Optimization TIPO to address this challenge. TIPO improves persona alignment and task executability, outperforming existing meth...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11259
• PDF: https://arxiv.org/pdf/2604.11259

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

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

#MobileAI #PrivacyTech #Personalization #GUIAgents #MachineLearning
SHARE: Social-Humanities AI for Research and Education

📝 Summary:
SHARE models are causal language models pre-trained specifically for social sciences and humanities that match general-purpose model performance while MIRROR provides a text review interface that pres...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11152
• PDF: https://arxiv.org/pdf/2604.11152
• Github: https://github.com/Joaoffg/SHARE

🔹 Models citing this paper:
https://huggingface.co/Joaoffg/SHARE-4B-Base-2604
https://huggingface.co/Joaoffg/SHARE-14B-Base-2604

Datasets citing this paper:
https://huggingface.co/datasets/Joaoffg/Cloze-SSH

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
SCOPE: Signal-Calibrated On-Policy Distillation Enhancement with Dual-Path Adaptive Weighting

📝 Summary:
SCOPE enhances on-policy distillation by adapting supervision paths based on trajectory correctness, using teacher-perplexity-weighted KL distillation for incorrect trajectories and student-perplexity...

🔹 Publication Date: Published on Apr 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10688
• PDF: https://arxiv.org/pdf/2604.10688

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Learning Long-term Motion Embeddings for Efficient Kinematics Generation

📝 Summary:
Efficient motion generation is achieved through compressed motion embeddings and conditional flow-matching models that produce realistic long-term motions from text prompts or spatial inputs. AI-gener...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11737
• PDF: https://arxiv.org/pdf/2604.11737
• Project Page: https://compvis.github.io/long-term-motion/
• Github: https://github.com/CompVis/long-term-motion

🔹 Models citing this paper:
https://huggingface.co/CompVis/ZipMo

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
How Alignment Routes: Localizing, Scaling, and Controlling Policy Circuits in Language Models

📝 Summary:
The study reveals that policy routing in alignment-trained language models involves attention gates and amplifier heads that control safety responses, with the routing mechanism being early-committing...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04385
• PDF: https://arxiv.org/pdf/2604.04385
• Github: https://github.com/gregfrank/how-alignment-routes

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Counting to Four is still a Chore for VLMs

📝 Summary:
Vision-language models exhibit counting failures due to reduced visual evidence utilization in later language layers, which can be mitigated through modality attention share interventions. AI-generate...

🔹 Publication Date: Published on Apr 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.10039
• PDF: https://arxiv.org/pdf/2604.10039
• Project Page: https://huggingface.co/papers?q=modality%20projection%20stage
• Github: https://github.com/leduy99/-CVPRW26-Modality-Attention-Share

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
1👍1
Panoptic Pairwise Distortion Graph

📝 Summary:
Researchers introduce a novel approach to image assessment by representing image pairs as structured distortion graphs that capture region-level degradation information, challenging existing multimoda...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11004
• PDF: https://arxiv.org/pdf/2604.11004
• Project Page: https://aismartperception.github.io/distortion-graph/
• Github: https://github.com/AISmartPerception/distortion-graphs

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Agentic Aggregation for Parallel Scaling of Long-Horizon Agentic Tasks

📝 Summary:
AggAgent enables efficient parallel test-time scaling for long-horizon agentic tasks by aggregating trajectories through a lightweight agent that navigates and synthesizes information on demand. AI-ge...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11753
• PDF: https://arxiv.org/pdf/2604.11753

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
1