ML Research Hub
32.8K subscribers
4.28K photos
258 videos
23 files
4.63K links
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.

Admin: @HusseinSheikho || @Hussein_Sheikho
Download Telegram
InfoSynth: Information-Guided Benchmark Synthesis for LLMs

📝 Summary:
InfoSynth automatically generates novel and diverse coding benchmarks for LLMs. It uses information-theoretic metrics and genetic algorithms to create scalable self-verifying problems, overcoming manual effort and training data contamination.

🔹 Publication Date: Published on Jan 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.00575
• PDF: https://arxiv.org/pdf/2601.00575
• Project Page: https://ishirgarg.github.io/infosynth_web/
• Github: https://github.com/ishirgarg/infosynth

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

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

#LLM #AI #Benchmarking #GenerativeAI #DeepLearning
Diversity or Precision? A Deep Dive into Next Token Prediction

📝 Summary:
This paper proposes a pre-training objective that reshapes the token-output distribution for better RL exploration. It uses reward-shaping to balance diversity and precision in next-token prediction. Contrary to intuition, a precision-oriented prior surprisingly yields a superior exploration spac...

🔹 Publication Date: Published on Dec 28, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.22955
• PDF: https://arxiv.org/pdf/2512.22955

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

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

#NextTokenPrediction #ReinforcementLearning #LLM #NLP #AIResearch
1
OmniVCus: Feedforward Subject-driven Video Customization with Multimodal Control Conditions

📝 Summary:
OmniVCus introduces a system for feedforward multi-subject video customization with multimodal controls. It proposes a data pipeline, VideoCus-Factory, and a diffusion Transformer framework with novel embedding mechanisms. This enables more subjects and precise editing, significantly outperformin...

🔹 Publication Date: Published on Jun 29, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2506.23361
• PDF: https://arxiv.org/pdf/2506.23361
• Project Page: https://caiyuanhao1998.github.io/project/OmniVCus/
• Github: https://github.com/caiyuanhao1998/Open-OmniVCus

🔹 Models citing this paper:
https://huggingface.co/CaiYuanhao/OmniVCus

Datasets citing this paper:
https://huggingface.co/datasets/CaiYuanhao/OmniVCus
https://huggingface.co/datasets/CaiYuanhao/OmniVCus-Test
https://huggingface.co/datasets/CaiYuanhao/OmniVCus-Train

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

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

#VideoGeneration #DiffusionModels #MultimodalAI #DeepLearning #ComputerVision
1
Media is too big
VIEW IN TELEGRAM
OnSpace Mobile App builder: Build AI Apps in minutes

Visit website: https://www.onspace.ai/?via=tg_datas
Or Download app:https://onspace.onelink.me/za8S/h1jb6sb9?c=datas

With OnSpace, you can build website or AI Mobile Apps by chatting with AI, and publish to PlayStore or AppStore.

What will you get:
✔️ Create app or website by chatting with AI;
✔️ Integrate with Any top AI power just by giving order (like Sora2, Nanobanan Pro & Gemini 3 Pro);
✔️ Download APK,AAB file, publish to AppStore.
✔️ Add payments and monetize like in-app-purchase and Stripe.
✔️ Functional login & signup.
✔️ Database + dashboard in minutes.
✔️ Full tutorial on YouTube and within 1 day customer service
Please open Telegram to view this post
VIEW IN TELEGRAM
2
Bitnet.cpp: Efficient Edge Inference for Ternary LLMs

📝 Summary:
Bitnet.cpp enhances edge inference for ternary LLMs using a novel mixed-precision matrix multiplication library. This system incorporates Ternary Lookup Tables and Int2 with a Scale for efficient, lossless inference, achieving up to a 6.25x speed increase over baselines.

🔹 Publication Date: Published on Feb 17, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2502.11880
• PDF: https://arxiv.org/pdf/2502.11880
• Github: https://github.com/microsoft/BitNet/tree/paper

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

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

#LLM #EdgeAI #MachineLearning #DeepLearning #AI
1
BitNet b1.58 2B4T Technical Report

📝 Summary:
BitNet b1.58 2B4T is the first open-source 1-bit Large Language Model with 2 billion parameters. It matches full-precision LLM performance while offering significant improvements in computational efficiency like reduced memory and energy. The model weights are openly released for research.

🔹 Publication Date: Published on Apr 16, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2504.12285
• PDF: https://arxiv.org/pdf/2504.12285
• Github: https://github.com/microsoft/bitnet

🔹 Models citing this paper:
https://huggingface.co/microsoft/bitnet-b1.58-2B-4T
https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf
https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-bf16

Spaces citing this paper:
https://huggingface.co/spaces/suayptalha/Chat-with-Bitnet-b1.58-2B-4T
https://huggingface.co/spaces/aizip-dev/SLM-RAG-Arena
https://huggingface.co/spaces/Tonic/Native_1-bit_LLM

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

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

#LLM #AI #Quantization #OpenSourceAI #DeepLearning
Taming Preference Mode Collapse via Directional Decoupling Alignment in Diffusion Reinforcement Learning

📝 Summary:
This paper addresses Preference Mode Collapse PMC in text-to-image diffusion models, where models lose diversity despite high reward scores. It introduces D^2-Align, a framework that mitigates PMC by directionally correcting the reward signal during optimization. This novel approach maintains gen...

🔹 Publication Date: Published on Dec 30, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.24146
• PDF: https://arxiv.org/pdf/2512.24146

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

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

#DiffusionModels #ReinforcementLearning #GenerativeAI #MachineLearning #AIResearch
This media is not supported in your browser
VIEW IN TELEGRAM
DreamID-V:Bridging the Image-to-Video Gap for High-Fidelity Face Swapping via Diffusion Transformer

📝 Summary:
DreamID-V is a novel video face swapping framework that uses diffusion transformers and curriculum learning. It achieves superior identity preservation and visual realism by bridging the image-to-video gap, outperforming existing methods and enhancing temporal consistency.

🔹 Publication Date: Published on Jan 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01425
• PDF: https://arxiv.org/pdf/2601.01425
• Project Page: https://guoxu1233.github.io/DreamID-V/
• Github: https://guoxu1233.github.io/DreamID-V/

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

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

#FaceSwapping #DiffusionModels #ComputerVision #GenerativeAI #VideoAI
BitNet Distillation

📝 Summary:
BitNet Distillation fine-tunes LLMs to 1.58-bit precision using SubLN, attention distillation, and continual pre-training. It achieves comparable performance to full-precision models, offering 10x memory savings and 2.65x faster inference.

🔹 Publication Date: Published on Oct 15, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2510.13998
• PDF: https://arxiv.org/pdf/2510.13998
• Github: https://github.com/microsoft/BitNet

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

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

#LLM #Quantization #ModelCompression #DeepLearning #AI
NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation

📝 Summary:
NextFlow is a unified decoder-only transformer enabling fast multimodal understanding and generation. It uses next-token prediction for text and next-scale for images, generating 1024x1024 images in 5 seconds. It achieves state-of-the-art performance among unified models.

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02204
• PDF: https://arxiv.org/pdf/2601.02204
• Github: https://github.com/ByteVisionLab/NextFlow

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Can LLMs Predict Their Own Failures? Self-Awareness via Internal Circuits

📝 Summary:
Large language models (LLMs) generate fluent and complex outputs but often fail to recognize their own mistakes and hallucinations. Existing approaches typically rely on external judges, multi-sample ...

🔹 Publication Date: Published on Dec 23, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.20578
• PDF: https://arxiv.org/pdf/2512.20578
• Github: https://github.com/Amirhosein-gh98/Gnosis

🔹 Models citing this paper:
https://huggingface.co/AmirhoseinGH/Gnosis-Qwen3-1.7B-Hybrid
https://huggingface.co/AmirhoseinGH/Gnosis-Qwen3-4B-Instruct-2507
https://huggingface.co/AmirhoseinGH/Gnosis-Qwen3-4B-Thinking-2507

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
VAR RL Done Right: Tackling Asynchronous Policy Conflicts in Visual Autoregressive Generation

📝 Summary:
Visual autoregressive models face training instability due to asynchronous policy conflicts, which are addressed through a novel framework enhancing group relative policy optimization with intermediat...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02256
• PDF: https://arxiv.org/pdf/2601.02256

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
This media is not supported in your browser
VIEW IN TELEGRAM
Talk2Move: Reinforcement Learning for Text-Instructed Object-Level Geometric Transformation in Scenes

📝 Summary:
Talk2Move presents a reinforcement learning-based diffusion framework that enables precise, semantically faithful spatial transformations of objects in scenes using natural language instructions. AI-g...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02356
• PDF: https://arxiv.org/pdf/2601.02356
• Project Page: https://sparkstj.github.io/talk2move/
• Github: https://github.com/sparkstj/Talk2Move

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
KV-Embedding: Training-free Text Embedding via Internal KV Re-routing in Decoder-only LLMs

📝 Summary:
KV-Embedding enables training-free representation learning from frozen LLMs by utilizing key-value states for enhanced context access and automated layer selection. AI-generated summary While LLMs are...

🔹 Publication Date: Published on Jan 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01046
• PDF: https://arxiv.org/pdf/2601.01046

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
VINO: A Unified Visual Generator with Interleaved OmniModal Context

📝 Summary:
VINO is a unified visual generator that uses a shared diffusion backbone with multimodal inputs to perform image and video generation and editing tasks. AI-generated summary We present VINO, a unified...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02358
• PDF: https://arxiv.org/pdf/2601.02358
• Project Page: https://sotamak1r.github.io/VINO-web/
• Github: https://github.com/SOTAMak1r/VINO-code

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
K-EXAONE Technical Report

📝 Summary:
K-EXAONE is a multilingual language model with a Mixture-of-Experts architecture that achieves competitive performance on various benchmarks while supporting multiple languages and long-context window...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01739
• PDF: https://arxiv.org/pdf/2601.01739
• Github: https://github.com/LG-AI-EXAONE/K-EXAONE

🔹 Models citing this paper:
https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling

📝 Summary:
Falcon-H1R is a 7B-parameter language model that achieves competitive reasoning performance through efficient training strategies and architectural design, enabling scalable reasoning capabilities in ...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02346
• PDF: https://arxiv.org/pdf/2601.02346

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
OpenNovelty: An LLM-powered Agentic System for Verifiable Scholarly Novelty Assessment

📝 Summary:
OpenNovelty is an LLM-powered agentic system for verifiable scholarly novelty assessment in peer review. It retrieves and analyzes prior work via semantic search and taxonomy construction, generating evidence-backed reports grounded in real papers. This tool aims to promote fair, consistent, and ...

🔹 Publication Date: Published on Jan 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01576
• PDF: https://arxiv.org/pdf/2601.01576
• Project Page: https://www.opennovelty.org/
• Github: https://github.com/january-blue/OpenNovelty

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research