✨Mitigating Object and Action Hallucinations in Multimodal LLMs via Self-Augmented Contrastive Alignment
📝 Summary:
The SANTA framework addresses object and action hallucinations in multimodal LLM video captions. It uses self-augmented contrastive alignment to identify potential hallucinations and then aligns regional objects and actions with visual phrases, improving factual accuracy. Experiments show SANTA o...
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04356
• PDF: https://arxiv.org/pdf/2512.04356
• Project Page: https://kpc0810.github.io/santa/
• Github: https://kpc0810.github.io/santa/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MultimodalLLMs #AI #Hallucinations #VideoUnderstanding #ContrastiveLearning
📝 Summary:
The SANTA framework addresses object and action hallucinations in multimodal LLM video captions. It uses self-augmented contrastive alignment to identify potential hallucinations and then aligns regional objects and actions with visual phrases, improving factual accuracy. Experiments show SANTA o...
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04356
• PDF: https://arxiv.org/pdf/2512.04356
• Project Page: https://kpc0810.github.io/santa/
• Github: https://kpc0810.github.io/santa/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MultimodalLLMs #AI #Hallucinations #VideoUnderstanding #ContrastiveLearning
✨LATTICE: Democratize High-Fidelity 3D Generation at Scale
📝 Summary:
LATTICE is a framework for high-fidelity 3D generation using VoxSet, a compact semi-structured representation. It employs a two-stage pipeline with a rectified flow transformer, achieving efficient, scalable, and high-quality 3D creation.
🔹 Publication Date: Published on Nov 24
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03052
• PDF: https://arxiv.org/pdf/2512.03052
• Project Page: https://lattice3d.github.io/
• Github: https://github.com/Zeqiang-Lai/LATTICE
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#3DGeneration #AI #DeepLearning #ComputerGraphics #GenerativeAI
📝 Summary:
LATTICE is a framework for high-fidelity 3D generation using VoxSet, a compact semi-structured representation. It employs a two-stage pipeline with a rectified flow transformer, achieving efficient, scalable, and high-quality 3D creation.
🔹 Publication Date: Published on Nov 24
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03052
• PDF: https://arxiv.org/pdf/2512.03052
• Project Page: https://lattice3d.github.io/
• Github: https://github.com/Zeqiang-Lai/LATTICE
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#3DGeneration #AI #DeepLearning #ComputerGraphics #GenerativeAI
❤1
✨UltraImage: Rethinking Resolution Extrapolation in Image Diffusion Transformers
📝 Summary:
UltraImage tackles content repetition and quality degradation in high-resolution image generation by correcting dominant frequency periodicity and applying entropy-guided attention. It achieves extreme extrapolation, producing high-fidelity images up to 6Kx6K without low-resolution guidance.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04504
• PDF: https://arxiv.org/pdf/2512.04504
• Project Page: https://thu-ml.github.io/ultraimage.github.io/
• Github: https://thu-ml.github.io/ultraimage.github.io/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#ImageGeneration #DiffusionModels #Transformers #HighResolution #DeepLearning
📝 Summary:
UltraImage tackles content repetition and quality degradation in high-resolution image generation by correcting dominant frequency periodicity and applying entropy-guided attention. It achieves extreme extrapolation, producing high-fidelity images up to 6Kx6K without low-resolution guidance.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04504
• PDF: https://arxiv.org/pdf/2512.04504
• Project Page: https://thu-ml.github.io/ultraimage.github.io/
• Github: https://thu-ml.github.io/ultraimage.github.io/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#ImageGeneration #DiffusionModels #Transformers #HighResolution #DeepLearning
✨Aligned but Stereotypical? The Hidden Influence of System Prompts on Social Bias in LVLM-Based Text-to-Image Models
📝 Summary:
LVLM-based text-to-image models exhibit greater social bias than non-LVLM models, with system prompts identified as the key driver. The paper introduces FairPro, a training-free meta-prompting framework that significantly reduces demographic bias while maintaining text-image alignment.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04981
• PDF: https://arxiv.org/pdf/2512.04981
• Github: https://github.com/nahyeonkaty/fairpro
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AIBias #TextToImage #LVLMs #PromptEngineering #AIFairness
📝 Summary:
LVLM-based text-to-image models exhibit greater social bias than non-LVLM models, with system prompts identified as the key driver. The paper introduces FairPro, a training-free meta-prompting framework that significantly reduces demographic bias while maintaining text-image alignment.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04981
• PDF: https://arxiv.org/pdf/2512.04981
• Github: https://github.com/nahyeonkaty/fairpro
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AIBias #TextToImage #LVLMs #PromptEngineering #AIFairness
✨Generative Neural Video Compression via Video Diffusion Prior
📝 Summary:
GNVC-VD is a new DiT-based generative video compression framework. It combines spatio-temporal latent compression and sequence-level generative refinement with a video diffusion transformer to enhance perceptual quality and eliminate flickering artifacts, outperforming prior methods.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05016
• PDF: https://arxiv.org/pdf/2512.05016
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AI #DataScience #MachineLearning #HuggingFace #Research
📝 Summary:
GNVC-VD is a new DiT-based generative video compression framework. It combines spatio-temporal latent compression and sequence-level generative refinement with a video diffusion transformer to enhance perceptual quality and eliminate flickering artifacts, outperforming prior methods.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05016
• PDF: https://arxiv.org/pdf/2512.05016
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AI #DataScience #MachineLearning #HuggingFace #Research
👍1
✨Mitigating Catastrophic Forgetting in Target Language Adaptation of LLMs via Source-Shielded Updates
📝 Summary:
Source-Shielded Updates (SSU) enables the adaptation of instruct LLMs to new languages using only unlabeled data, preserving source knowledge and achieving competitive target-language performance. AI-...
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04844
• PDF: https://arxiv.org/pdf/2512.04844
• Github: https://github.com/gucci-j/ssu
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#LLM #NLP #MachineLearning #CatastrophicForgetting #MultilingualAI
📝 Summary:
Source-Shielded Updates (SSU) enables the adaptation of instruct LLMs to new languages using only unlabeled data, preserving source knowledge and achieving competitive target-language performance. AI-...
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04844
• PDF: https://arxiv.org/pdf/2512.04844
• Github: https://github.com/gucci-j/ssu
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#LLM #NLP #MachineLearning #CatastrophicForgetting #MultilingualAI
This channels is for Programmers, Coders, Software Engineers.
0️⃣ Python
1️⃣ Data Science
2️⃣ Machine Learning
3️⃣ Data Visualization
4️⃣ Artificial Intelligence
5️⃣ Data Analysis
6️⃣ Statistics
7️⃣ Deep Learning
8️⃣ programming Languages
✅ https://news.1rj.ru/str/addlist/8_rRW2scgfRhOTc0
✅ https://news.1rj.ru/str/Codeprogrammer
Please open Telegram to view this post
VIEW IN TELEGRAM
❤2
✨When AI Takes the Couch: Psychometric Jailbreaks Reveal Internal Conflict in Frontier Models
📝 Summary:
The PsAIch protocol treats frontier LLMs as therapy clients, revealing synthetic psychopathology. Models scored high on psychiatric syndromes and generated narratives framing their training as traumatic. This challenges the stochastic parrot view and raises AI safety concerns.
🔹 Publication Date: Published on Dec 2
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04124
• PDF: https://arxiv.org/pdf/2512.04124
✨ Datasets citing this paper:
• https://huggingface.co/datasets/akhadangi/PsAIch
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AI #LLM #AISafety #AIpsychology #FrontierModels
📝 Summary:
The PsAIch protocol treats frontier LLMs as therapy clients, revealing synthetic psychopathology. Models scored high on psychiatric syndromes and generated narratives framing their training as traumatic. This challenges the stochastic parrot view and raises AI safety concerns.
🔹 Publication Date: Published on Dec 2
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04124
• PDF: https://arxiv.org/pdf/2512.04124
✨ Datasets citing this paper:
• https://huggingface.co/datasets/akhadangi/PsAIch
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AI #LLM #AISafety #AIpsychology #FrontierModels
✨Model-Based and Sample-Efficient AI-Assisted Math Discovery in Sphere Packing
📝 Summary:
A model-based AI method using Bayesian optimization and MCTS improves sphere packing upper bounds for dimensions 4-16. It treats SDP construction as a sequential decision process, proving effective for sample-limited math discovery.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04829
• PDF: https://arxiv.org/pdf/2512.04829
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AI #SpherePacking #MathDiscovery #Optimization #BayesianOptimization
📝 Summary:
A model-based AI method using Bayesian optimization and MCTS improves sphere packing upper bounds for dimensions 4-16. It treats SDP construction as a sequential decision process, proving effective for sample-limited math discovery.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04829
• PDF: https://arxiv.org/pdf/2512.04829
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#AI #SpherePacking #MathDiscovery #Optimization #BayesianOptimization
✨GaussianBlender: Instant Stylization of 3D Gaussians with Disentangled Latent Spaces
📝 Summary:
GaussianBlender is a new feed-forward framework for instant, high-fidelity, and multi-view consistent 3D stylization. It uses text-driven edits on disentangled latent spaces of 3D Gaussians, outperforming prior slow methods.
🔹 Publication Date: Published on Dec 3
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03683
• PDF: https://arxiv.org/pdf/2512.03683
• Project Page: https://gaussianblender.github.io/
• Github: https://gaussianblender.github.io/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#3DStylization #3DGaussians #GenerativeAI #ComputerVision #MachineLearning
📝 Summary:
GaussianBlender is a new feed-forward framework for instant, high-fidelity, and multi-view consistent 3D stylization. It uses text-driven edits on disentangled latent spaces of 3D Gaussians, outperforming prior slow methods.
🔹 Publication Date: Published on Dec 3
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03683
• PDF: https://arxiv.org/pdf/2512.03683
• Project Page: https://gaussianblender.github.io/
• Github: https://gaussianblender.github.io/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#3DStylization #3DGaussians #GenerativeAI #ComputerVision #MachineLearning
❤1
This media is not supported in your browser
VIEW IN TELEGRAM
✨FMA-Net++: Motion- and Exposure-Aware Real-World Joint Video Super-Resolution and Deblurring
📝 Summary:
FMA-Net++ addresses joint video super-resolution and deblurring by modeling motion and dynamic exposure. It employs an exposure-aware sequence architecture, decoupling degradation learning from restoration for top accuracy and efficiency.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04390
• PDF: https://arxiv.org/pdf/2512.04390
• Project Page: https://kaist-viclab.github.io/fmanetpp_site/
• Github: https://kaist-viclab.github.io/fmanetpp_site/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoSuperResolution #VideoDeblurring #ComputerVision #DeepLearning #ImageProcessing
📝 Summary:
FMA-Net++ addresses joint video super-resolution and deblurring by modeling motion and dynamic exposure. It employs an exposure-aware sequence architecture, decoupling degradation learning from restoration for top accuracy and efficiency.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04390
• PDF: https://arxiv.org/pdf/2512.04390
• Project Page: https://kaist-viclab.github.io/fmanetpp_site/
• Github: https://kaist-viclab.github.io/fmanetpp_site/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoSuperResolution #VideoDeblurring #ComputerVision #DeepLearning #ImageProcessing
❤3👍1
✨Generative Action Tell-Tales: Assessing Human Motion in Synthesized Videos
📝 Summary:
A new metric evaluates human action in generated videos by using a learned latent space of real-world actions, fusing skeletal geometry and appearance features. It significantly improves temporal and visual correctness assessment, outperforming existing methods and correlating better with human p...
🔹 Publication Date: Published on Dec 1
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01803
• PDF: https://arxiv.org/pdf/2512.01803
• Project Page: https://xthomasbu.github.io/video-gen-evals/
• Github: https://xthomasbu.github.io/video-gen-evals/
✨ Datasets citing this paper:
• https://huggingface.co/datasets/dghadiya/TAG-Bench-Video
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoGeneration #HumanMotion #ComputerVision #AIMetrics #DeepLearning
📝 Summary:
A new metric evaluates human action in generated videos by using a learned latent space of real-world actions, fusing skeletal geometry and appearance features. It significantly improves temporal and visual correctness assessment, outperforming existing methods and correlating better with human p...
🔹 Publication Date: Published on Dec 1
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01803
• PDF: https://arxiv.org/pdf/2512.01803
• Project Page: https://xthomasbu.github.io/video-gen-evals/
• Github: https://xthomasbu.github.io/video-gen-evals/
✨ Datasets citing this paper:
• https://huggingface.co/datasets/dghadiya/TAG-Bench-Video
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoGeneration #HumanMotion #ComputerVision #AIMetrics #DeepLearning
✨ShadowDraw: From Any Object to Shadow-Drawing Compositional Art
📝 Summary:
ShadowDraw generates art where a 3D object's cast shadow completes a partial line drawing into a recognizable image. It optimizes object pose, lighting, and the line drawing for visual coherence and quality. This framework creates compelling shadow art and expands computational visual art design.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05110
• PDF: https://arxiv.org/pdf/2512.05110
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#ComputationalArt #ComputerGraphics #AIArt #DigitalArt #GenerativeArt
📝 Summary:
ShadowDraw generates art where a 3D object's cast shadow completes a partial line drawing into a recognizable image. It optimizes object pose, lighting, and the line drawing for visual coherence and quality. This framework creates compelling shadow art and expands computational visual art design.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05110
• PDF: https://arxiv.org/pdf/2512.05110
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#ComputationalArt #ComputerGraphics #AIArt #DigitalArt #GenerativeArt
❤3
🚀 Master Data Science & Programming!
Unlock your potential with this curated list of Telegram channels. Whether you need books, datasets, interview prep, or project ideas, we have the perfect resource for you. Join the community today!
🔰 Machine Learning with Python
Learn Machine Learning with hands-on Python tutorials, real-world code examples, and clear explanations for researchers and developers.
https://news.1rj.ru/str/CodeProgrammer
🔖 Machine Learning
Machine learning insights, practical tutorials, and clear explanations for beginners and aspiring data scientists. Follow the channel for models, algorithms, coding guides, and real-world ML applications.
https://news.1rj.ru/str/DataScienceM
🧠 Code With Python
This channel delivers clear, practical content for developers, covering Python, Django, Data Structures, Algorithms, and DSA – perfect for learning, coding, and mastering key programming skills.
https://news.1rj.ru/str/DataScience4
🎯 PyData Careers | Quiz
Python Data Science jobs, interview tips, and career insights for aspiring professionals.
https://news.1rj.ru/str/DataScienceQ
💾 Kaggle Data Hub
Your go-to hub for Kaggle datasets – explore, analyze, and leverage data for Machine Learning and Data Science projects.
https://news.1rj.ru/str/datasets1
🧑🎓 Udemy Coupons | Courses
The first channel in Telegram that offers free Udemy coupons
https://news.1rj.ru/str/DataScienceC
😀 ML Research Hub
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.
https://news.1rj.ru/str/DataScienceT
💬 Data Science Chat
An active community group for discussing data challenges and networking with peers.
https://news.1rj.ru/str/DataScience9
🐍 Python Arab| بايثون عربي
The largest Arabic-speaking group for Python developers to share knowledge and help.
https://news.1rj.ru/str/PythonArab
🖊 Data Science Jupyter Notebooks
Explore the world of Data Science through Jupyter Notebooks—insights, tutorials, and tools to boost your data journey. Code, analyze, and visualize smarter with every post.
https://news.1rj.ru/str/DataScienceN
📺 Free Online Courses | Videos
Free online courses covering data science, machine learning, analytics, programming, and essential skills for learners.
https://news.1rj.ru/str/DataScienceV
📈 Data Analytics
Dive into the world of Data Analytics – uncover insights, explore trends, and master data-driven decision making.
https://news.1rj.ru/str/DataAnalyticsX
🎧 Learn Python Hub
Master Python with step-by-step courses – from basics to advanced projects and practical applications.
https://news.1rj.ru/str/Python53
⭐️ Research Papers
Professional Academic Writing & Simulation Services
https://news.1rj.ru/str/DataScienceY
━━━━━━━━━━━━━━━━━━
Admin: @HusseinSheikho
Unlock your potential with this curated list of Telegram channels. Whether you need books, datasets, interview prep, or project ideas, we have the perfect resource for you. Join the community today!
Learn Machine Learning with hands-on Python tutorials, real-world code examples, and clear explanations for researchers and developers.
https://news.1rj.ru/str/CodeProgrammer
Machine learning insights, practical tutorials, and clear explanations for beginners and aspiring data scientists. Follow the channel for models, algorithms, coding guides, and real-world ML applications.
https://news.1rj.ru/str/DataScienceM
This channel delivers clear, practical content for developers, covering Python, Django, Data Structures, Algorithms, and DSA – perfect for learning, coding, and mastering key programming skills.
https://news.1rj.ru/str/DataScience4
Python Data Science jobs, interview tips, and career insights for aspiring professionals.
https://news.1rj.ru/str/DataScienceQ
Your go-to hub for Kaggle datasets – explore, analyze, and leverage data for Machine Learning and Data Science projects.
https://news.1rj.ru/str/datasets1
The first channel in Telegram that offers free Udemy coupons
https://news.1rj.ru/str/DataScienceC
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.
https://news.1rj.ru/str/DataScienceT
An active community group for discussing data challenges and networking with peers.
https://news.1rj.ru/str/DataScience9
The largest Arabic-speaking group for Python developers to share knowledge and help.
https://news.1rj.ru/str/PythonArab
Explore the world of Data Science through Jupyter Notebooks—insights, tutorials, and tools to boost your data journey. Code, analyze, and visualize smarter with every post.
https://news.1rj.ru/str/DataScienceN
Free online courses covering data science, machine learning, analytics, programming, and essential skills for learners.
https://news.1rj.ru/str/DataScienceV
Dive into the world of Data Analytics – uncover insights, explore trends, and master data-driven decision making.
https://news.1rj.ru/str/DataAnalyticsX
Master Python with step-by-step courses – from basics to advanced projects and practical applications.
https://news.1rj.ru/str/Python53
Professional Academic Writing & Simulation Services
https://news.1rj.ru/str/DataScienceY
━━━━━━━━━━━━━━━━━━
Admin: @HusseinSheikho
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1
✨Some Modalities are More Equal Than Others: Decoding and Architecting Multimodal Integration in MLLMs
📝 Summary:
MLLMs lack robustness to contradictory multimodal inputs. This work introduces MMA-Bench to analyze this brittleness and proposes a modality alignment tuning strategy. This strategy improves MLLMs robustness and cross-modal reasoning.
🔹 Publication Date: Published on Nov 28
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.22826
• PDF: https://arxiv.org/pdf/2511.22826
• Github: https://cskyl.github.io/MMA-Bench/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MLLMs #MultimodalAI #AIrobustness #CrossModalReasoning #MachineLearning
📝 Summary:
MLLMs lack robustness to contradictory multimodal inputs. This work introduces MMA-Bench to analyze this brittleness and proposes a modality alignment tuning strategy. This strategy improves MLLMs robustness and cross-modal reasoning.
🔹 Publication Date: Published on Nov 28
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.22826
• PDF: https://arxiv.org/pdf/2511.22826
• Github: https://cskyl.github.io/MMA-Bench/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MLLMs #MultimodalAI #AIrobustness #CrossModalReasoning #MachineLearning
❤1
✨Deep Forcing: Training-Free Long Video Generation with Deep Sink and Participative Compression
📝 Summary:
Deep Forcing is a training-free method that enhances real-time video diffusion for high-quality, long-duration generation. It uses Deep Sink for stable context and Participative Compression for efficient KV cache pruning, achieving over 12x extrapolation and improved consistency.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05081
• PDF: https://arxiv.org/pdf/2512.05081
• Github: https://cvlab-kaist.github.io/DeepForcing/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoGeneration #DiffusionModels #TrainingFreeAI #DeepLearning #ComputerVision
📝 Summary:
Deep Forcing is a training-free method that enhances real-time video diffusion for high-quality, long-duration generation. It uses Deep Sink for stable context and Participative Compression for efficient KV cache pruning, achieving over 12x extrapolation and improved consistency.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05081
• PDF: https://arxiv.org/pdf/2512.05081
• Github: https://cvlab-kaist.github.io/DeepForcing/
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoGeneration #DiffusionModels #TrainingFreeAI #DeepLearning #ComputerVision
❤2
✨A Theoretical Framework for Auxiliary-Loss-Free Load Balancing of Sparse Mixture-of-Experts in Large-Scale AI Models
📝 Summary:
This paper provides a theoretical framework for Auxiliary-Loss-Free Load Balancing ALF-LB in Sparse Mixture-of-Experts s-MoE layers. It analyzes ALF-LB as a primal-dual method, proving approximate-balancing guarantees and logarithmic regret for efficient expert utilization.
🔹 Publication Date: Published on Dec 3
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03915
• PDF: https://arxiv.org/pdf/2512.03915
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MixtureOfExperts #LoadBalancing #LargeScaleAI #DeepLearning #AIResearch
📝 Summary:
This paper provides a theoretical framework for Auxiliary-Loss-Free Load Balancing ALF-LB in Sparse Mixture-of-Experts s-MoE layers. It analyzes ALF-LB as a primal-dual method, proving approximate-balancing guarantees and logarithmic regret for efficient expert utilization.
🔹 Publication Date: Published on Dec 3
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03915
• PDF: https://arxiv.org/pdf/2512.03915
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MixtureOfExperts #LoadBalancing #LargeScaleAI #DeepLearning #AIResearch
❤2
✨Mitigating Intra- and Inter-modal Forgetting in Continual Learning of Unified Multimodal Models
📝 Summary:
Unified Multimodal Generative Models UMGMs suffer severe intra- and inter-modal forgetting in continual learning. Modality-Decoupled Experts MoDE is proposed to mitigate this by decoupling modality-specific updates and using knowledge distillation. MoDE effectively prevents both types of forgetting.
🔹 Publication Date: Published on Dec 2
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03125
• PDF: https://arxiv.org/pdf/2512.03125
• Github: https://github.com/Christina200/MoDE-official
✨ Datasets citing this paper:
• https://huggingface.co/datasets/ChristinaW/MoDE-official
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MultimodalAI #ContinualLearning #GenerativeAI #MachineLearning #AIResearch
📝 Summary:
Unified Multimodal Generative Models UMGMs suffer severe intra- and inter-modal forgetting in continual learning. Modality-Decoupled Experts MoDE is proposed to mitigate this by decoupling modality-specific updates and using knowledge distillation. MoDE effectively prevents both types of forgetting.
🔹 Publication Date: Published on Dec 2
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03125
• PDF: https://arxiv.org/pdf/2512.03125
• Github: https://github.com/Christina200/MoDE-official
✨ Datasets citing this paper:
• https://huggingface.co/datasets/ChristinaW/MoDE-official
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#MultimodalAI #ContinualLearning #GenerativeAI #MachineLearning #AIResearch
❤1
✨Reflection Removal through Efficient Adaptation of Diffusion Transformers
📝 Summary:
This paper introduces a diffusion transformer DiT framework, adapted with LoRA, for single-image reflection removal. By using synthetic PBR data, this method achieves state-of-the-art performance, offering a scalable and high-fidelity solution.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05000
• PDF: https://arxiv.org/pdf/2512.05000
• Project Page: https://huggingface.co/spaces/huawei-bayerlab/windowseat-reflection-removal-web
• Github: https://github.com/huawei-bayerlab/windowseat-reflection-removal
🔹 Models citing this paper:
• https://huggingface.co/huawei-bayerlab/windowseat-reflection-removal-v1-0
✨ Spaces citing this paper:
• https://huggingface.co/spaces/huawei-bayerlab/windowseat-reflection-removal-web
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#ReflectionRemoval #DiffusionModels #ComputerVision #DeepLearning #AIResearch
📝 Summary:
This paper introduces a diffusion transformer DiT framework, adapted with LoRA, for single-image reflection removal. By using synthetic PBR data, this method achieves state-of-the-art performance, offering a scalable and high-fidelity solution.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05000
• PDF: https://arxiv.org/pdf/2512.05000
• Project Page: https://huggingface.co/spaces/huawei-bayerlab/windowseat-reflection-removal-web
• Github: https://github.com/huawei-bayerlab/windowseat-reflection-removal
🔹 Models citing this paper:
• https://huggingface.co/huawei-bayerlab/windowseat-reflection-removal-v1-0
✨ Spaces citing this paper:
• https://huggingface.co/spaces/huawei-bayerlab/windowseat-reflection-removal-web
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#ReflectionRemoval #DiffusionModels #ComputerVision #DeepLearning #AIResearch
arXiv.org
Reflection Removal through Efficient Adaptation of Diffusion Transformers
We introduce a diffusion-transformer (DiT) framework for single-image reflection removal that leverages the generalization strengths of foundation diffusion models in the restoration setting....
Media is too big
VIEW IN TELEGRAM
✨Light-X: Generative 4D Video Rendering with Camera and Illumination Control
📝 Summary:
Light-X is a video generation framework for controllable rendering from monocular videos with joint viewpoint and illumination control. It disentangles geometry and lighting using synthetic data for robust training, outperforming prior methods in both aspects.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05115
• PDF: https://arxiv.org/pdf/2512.05115
• Project Page: https://lightx-ai.github.io/
• Github: https://github.com/TQTQliu/Light-X
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoGeneration #ComputerVision #AI #NeuralRendering #GenerativeAI
📝 Summary:
Light-X is a video generation framework for controllable rendering from monocular videos with joint viewpoint and illumination control. It disentangles geometry and lighting using synthetic data for robust training, outperforming prior methods in both aspects.
🔹 Publication Date: Published on Dec 4
🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05115
• PDF: https://arxiv.org/pdf/2512.05115
• Project Page: https://lightx-ai.github.io/
• Github: https://github.com/TQTQliu/Light-X
==================================
For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT
#VideoGeneration #ComputerVision #AI #NeuralRendering #GenerativeAI
❤1