Pramit Saha

🚀First step towards #agentic #automation in #Federated #Learning Excited to share our latest work #FedAgentBench on automating Real-world Federated Learning with LLM Agents: arxiv.org/pdf/2509.23803 Work done in @UniofOxford #FederatedLearning #LLM #AIagent #Multiagentsystem

Announcing the #ICML2026 policy for self-ranking in reviews! 1. Authors rank their submissions 2. Reviews are submitted 3. The "Isotonic Mechanism" is run on rankings and review scores 4. Large discrepancies are flagged to ACs and SAC CC @weijie444 Read more for details 👇

MICCAI Conference to be Included into the ELLIS Recognised Publication Venues - Sign the Petition! c.org/D4dHY6TPRk via @UKChange

MICCAI Conference to be Included into the ELLIS Recognised Publication Venues - Sign the Petition! c.org/ytsVfvCL7v via @UKChange

1. Trust the Pakistanis to make a pigsty even out of the @OxfordUnion. And as always, they are genetically incapable of being truthful. So here's the complete story of how this so-called debate played out.

@Tanmoy_Chak I think big companies and reputed people in the community should pave the way. But all I see after the acceptance deadline is: Pleased to announce x number of papers accepted from our lab in Neurips/ICLR/etc. And of course companies considering it mandatory for recruitment!

The ICLR'26 reviewer-name leak is a reminder that we're no longer advancing science -- we're trapped in a system that often disrespects it. Maybe it's time for AI/ML to pause and rethink the entire conference culture. Wild idea: Stop all major AI/ML conferences for 1-2 years. Let papers live on arXiv, and be submitted to journals. Meanwhile, conferences rebuild real policies and accountability. Will this ever happen? Probably not. The incentives are too big. Its about MONEY and PROFIT. But the community deserves better and we need the courage to imagine alternatives. @iclr_conf @NeurIPSConf @RealAAAI

@Frankleo61 @GetActionModel I think that's a very interesting question. Do you have any ideas?

@PramitSaha5 Incredible to see LLM agents handle FL end-to-end! Makes me think how #LAM could help agents learn optimal actions in these workflows, not just follow instructions. @GetActionModel

🚀First step towards #agentic #automation in #Federated #Learning Excited to share our latest work #FedAgentBench on automating Real-world Federated Learning with LLM Agents: arxiv.org/pdf/2509.23803 Work done in @UniofOxford #FederatedLearning #LLM #AIagent #Multiagentsystem

🌍 Implication: A future where federated learning runs itself, coordinated by specialized LLM agents — scalable, privacy-preserving, and adaptive across domains. 🤝 Happy to collaborate! If you work in #federated #learning, #LLMs, or #agentic #systems — let’s talk!

Key Insights: 🧩Frontier models (GPT-4.1, DeepSeek-V3) excel at structured FL tasks—config writing, client selection, preprocessing 🧠 Still weak on implicit reasoning: label harmonization, schema alignment 📈 Bigger ≠ better→ multi-agent reasoning & domain grounding are key

🚀 Excited to announce our paper, “Self-supervised Normality Learning and Divergence Vector-guided Model Merging for Zero-shot Congenital Heart Disease Detection in Fetal Ultrasound Videos,” from the @NobleLabOxford , has been accepted to #MICCAI2025!

@NobleLabOxford @FetalUltrasound @OlgaPatey @Perceptron97 @y_m_asano 📰 Dive into the details: arxiv.org/pdf/2503.07799 This is true one-shot federated learning for real-world video understanding - a leap forward for collaborative, privacy-preserving medical AI. Could this be the future of multi-institutional healthcare AI? We think so! 🚀 (n/n)

@NobleLabOxford @FetalUltrasound @OlgaPatey @Perceptron97 @y_m_asano Why does this matter? CHD labels are scarce and privacy laws block data sharing. Our approach, tested on real data from 5 hospitals, boosts accuracy by 20% and F1-score by 54.6% -outperforming SOTA and centralized models on unseen test sets! (3/n)

🚀 Thrilled to announce our paper, “Self-supervised Normality Learning and Divergence Vector-guided Model Merging for Zero-shot Congenital Heart Disease Detection in Fetal Ultrasound Videos,” is accepted at #MICCAI2025! 🥳 (1/n)

One of the biggest bottlenecks in deploying visual AI and computer vision is annotation, which can be both costly and time-consuming. Today, we’re introducing Verified Auto Labeling, a new approach to AI-assisted annotation that achieves up to 95% of human-level performance while cutting labeling costs by up to 100,000x and time by 5,000x. Read the full paper: arxiv.org/abs/2506.02359

🔗 𝗪𝗵𝘆 𝗱𝗼𝗲𝘀 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿? Foundation models are massive, but medical & sensitive domains need private, custom, efficient adaptation. F³OCUS is first to jointly optimize for client needs & collective diversity in FM finetuning. #CVPR2025 #FederatedLearning #VLM

3️⃣ 𝗡𝗲𝘄 𝗩𝗤𝗔 𝗱𝗮𝘁𝗮𝘀𝗲𝘁: 707k+ VQA triplets, 9 modality-specific clients, 12 anatomical categories: Colon, Lung, Skin, Eye, Breast, Kidney, Blood, Femur, Heart, Liver, Pancreas, Spleen #MedAI #Computer #Vision

🚨 #CVPR2025 #Highlights from #Oxford #Noble #Group 𝗣𝗼𝘀𝘁𝗲𝗿 𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝟰, Saturday, #401, 4 PM! How would you 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗠𝗼𝗱𝗲𝗹𝘀 with small datasets & limited compute? Our #CVPR2025 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀 paper looks at this challenge! 👇

✨𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: 1️⃣𝗟𝗮𝘆𝗲𝗿 𝘀𝗲𝗹𝗲𝗰𝘁𝗶𝗼𝗻 using Neural Tangent Kernel: Maximizes client-specific learning by identifying crucial layers for tuning. 2️⃣𝗝𝗼𝗶𝗻𝘁 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻: Meta-heuristics ensure balanced, diverse, and efficient layer selection.

🔍 𝗪𝗵𝗮𝘁’𝘀 𝗻𝗲𝘄? We propose a new approach to fine-tuning large Vision-Language Models (VLMs) on resource-constrained clients in Federated Learning—essential for healthcare, where privacy matters most. #FederatedLearning #VisionLanguage #FoundationModels