Feryal Behbahani

Join us on the 10th of March (in person or virtually) for "Mobile phones, rapid tests, and AI; to Antimicrobial Resistance and beyond" I will be sharing some of our work behind a prototype developed at the Digital Health Hub for Antimicrobial Resistance at UCL, exploring how AI on smartphones could support point-of-care diagnostics, along with the architectural, technical, and UX considerations behind the implementation. It should be relevant to RSEs, data scientists, researchers interested in digital health, as well as health practitioners with interest in technology. It would be great to connect with others working in this domain.

6

📣🚴 Very excited to announce that our paper 'The Oxford RobotCycle Project: A Multimodal Urban Cycling Dataset for Assessing the Safety of Vulnerable Road Users' has been published in IEEE Transactions on Field Robotics (T-FR)! 📄 Paper: https://lnkd.in/eCWujWD4 This article presents the complete RobotCycle dataset, detailing the final sensing setup and backpack design, the integration of eye-gaze tracking glasses, and results on traffic and risk analysis, pose estimation and mapping, semantic segmentation and knowledge graphs, eye-gaze attention patterns, and more.. Huge thanks to my amazing co-authors, Divya Thuremella, Jumana B., Samuel Sze, Lanke Frank Tarimo Fu, Benjamin Hardin, Tyler Reinmund, Tobit Flatscher, Daniel Marques, Chris Prahacs, Lars Kunze, and Daniele De Martini, as well as our fantastic cyclists and everyone who participated in the preliminary phases of the project. We are also very grateful for the support we received from the University of Oxford, Department of Engineering Science, the Oxfordshire County Council, and the Department for Transport (DfT), United Kingdom. It’s been a long journey putting this work together, and we’re really excited to share it with the community. Feel free to reach out if you’d like to discuss or collaborate! 🚀

161

16 Comments

I’m excited to share that our scoping review preprint has been published on medRxiv: A Scoping Review of Algorithmic Equity, Data Diversity, and Inclusive Design in the Transformer Era of Clinical NLP https://lnkd.in/eWvWM8h6 Many thanks to Abeer Badawi, Ph.D., Elham Dolatabadi, and Farah Ahmad, MBBS, MPH, PhD, for their collaboration, guidance, and support throughout this work. The rapid digitization of healthcare has positioned transformer-based NLP models as powerful tools for managing clinical text. However, their growing integration into practice raises important and unresolved questions about equity and inclusivity. This review synthesizes 56 studies (2017–2024) and examines how equity is addressed across three dimensions: ➡️ Algorithmic equity — fairness audits are largely post hoc and fragmented ➡️ Data diversity & representativeness — persistent underrepresentation creates what we define as Data Diversity Debt ➡️ Participatory design — observed in only 11% of studies, revealing a major gap beyond clinician involvement. To move beyond descriptive audits, we propose an equity-by-design roadmap to embed fairness, inclusivity, and accountability across the full lifecycle of clinical NLP systems—and to retire Data Diversity Debt. #HealthAI #ClinicalNLP #AlgorithmicEquity #DigitalHealthEquity #AIethics #ResponsibleAI #medRxiv

26

2 Comments

I am very excited to share our latest publication on the arXiv! The ATLAS collaboration has just released our latest measurements of production cross-sections of W-boson pairs 🥳🎉 This work presents precise measurements of WW production in proton-proton collisions at 13 TeV, using 140 inverse fb of ATLAS data collected between 2015 and 2018. Huge efforts were devoted to derive precise data-driven background estimations, enabling a measurement in a fully jet-inclusive phase space. We achieve a fiducial cross-section measurement with just 3.1% uncertainty, the most precise ever performed in a hadron collider to date. Differential cross-section measurements show excellent agreement with the state-of-the-art theoretical predictions at the same level of precision, providing important tests of the strong and electroweak sectors of the Standard Model. These have been used to constrain anomalous interactions in the framework of the Standard Model effective field theory. Check out the full pre-print for details: https://lnkd.in/e69Bhg6k Proud to contribute to this endeavor within the ATLAS collaboration and so grateful to all colleagues involved! #ATLAS #CERN #LHC #HighEnergyPhysics #arXiv #ParticlePhysics #WW #StandardModel #SM #Electroweak #Diboson

16

Excited to share our #CVPR26 paper: HistoSelect! 🚀 We tackle the challenge of using LLMs for Whole-Slide Image (WSI) understanding. Standard LLMs rely on full attention, making them computationally expensive and difficult to interpret for pathology. Our approach: Mimic a pathologist’s workflow, grounding our model in the Information Bottleneck principle. - Semantic Stratification: We use a vision-language foundation model to unsupervisedly partition tissue into meaningful semantic regions. - Adaptive Sampling: Adaptively sample the visual tokens that are relevant to the specific diagnostic question. The result: A more efficient, interpretable, and principled framework that keeps only the information that truly matters for the diagnosis. Check out the full paper here: https://lnkd.in/gzQ9Wh_P This is a collaboration with Mayo Clinic, Stony Brook University, Harvard Medical and Stanford University’s. Congrats to Wentao Huang and all other collaborators.

31

1 Comment

Building a Production-Ready RAG System for Persian Language Processing I’m excited to share insights from developing an advanced Retrieval-Augmented Generation system specifically engineered for Persian/Farsi document processing, leveraging cutting-edge ML infrastructure and architectural patterns. 🔬 Technical Architecture Highlights: Advanced RAG Implementation: ∙ Self-Corrective Agentic RAG with adaptive query rewriting and iterative refinement ∙ Hybrid search architecture combining dense embeddings with SPLADE sparse representations ∙ Multi-agent orchestration enabling context-aware document interaction and general conversation modes Production Infrastructure: ∙ NVIDIA Triton Inference Server deployment on H100 GPUs with vLLM backend ∙ Milvus vector database for high-performance similarity search ∙ PostgreSQL for structured data with MinIO object storage for scalable file management ∙ Model ensemble: Qwen2.5-72B-Instruct-AWQ, DeepSeek-OCR, and sentence-transformers ML Pipeline Innovations: ∙ Automated synthetic QA pair generation using LangGraph workflows ∙ Cross-encoder reranking (ms-marco-MiniLM-L6-v2) for precision optimization ∙ Dynamic prompt management system with A/B testing capabilities and versioning ∙ Fine-tuning pipeline with quality-scored synthetic datasets Engineering Principles: ∙ Clean Architecture with dependency injection and separation of concerns ∙ Docker containerization for reproducible deployments ∙ Comprehensive observability with metrics tracking and performance monitoring This project demonstrates how to architect enterprise-grade RAG systems that combine theoretical ML advancements with practical production considerations, particularly for low-resource languages. Open to discussing technical implementation details, architectural decisions, or challenges in deploying LLMs at scale. #MLEngineering #RAG #NLP #LLMs #VectorSearch #ProductionML #AI #DeepLearning #Persian #TritonInference #CleanArchitecture

1

I’m excited to share that our work, "FaMTEB: Massive Text Embedding Benchmark in Persian Language", has been accepted to EMNLP 2025 (Findings)! 🚀 The lack of a comprehensive benchmark for evaluating text embeddings in Persian has been a major challenge. Our team at MCINEXT, with the goal of addressing this gap, introduced FaMTEB, a benchmark for evaluating text embeddings in Persian, which is now featured on the reputable MTEB leaderboard. I’m grateful to have collaborated with Erfan Zeinivand and Mehran Sarmadi on this project. A special thanks to everyone who shared their knowledge and offered support throughout this journey. 🔗 Explore the leaderboard: https://lnkd.in/dHfmhzW7 📄 Read our paper: https://lnkd.in/dwWVbsD4 #EMNLP #EMNLP2025 #MCINEXT #FaMTEB #MTEB #PersianNLP

16

Pleased to share our preview on a recent cross-attention graph neural network-based model EZSpecificity, a framework built on a curated enzyme-substrate database and integrating both sequence and structural information for predicting enzyme substrate specificity at scale. (Cui, H., Su, Y., Dean, T.J. et al. Enzyme specificity prediction using cross-attention graph neural networks. Nature 647, 639–647 (2025). https://lnkd.in/ekcTBgm5) The integrated datasets and machine-learning tools developed in this work could be extended to other enzyme families and adapted for chiral-product prediction, for example, through refined encoding protocols and dynamic structural representations. Such advances may further enhance selectivity and predictive accuracy at finer Enzyme Commission (EC) classification levels, thereby facilitating the efficient exploitation of enzymes for applications in synthesis and medicine. 50 day’s of free access to our article from 👇 https://lnkd.in/e6ZYn_9i Chemistry and Chemical Engineering at Queen's Queen's University Belfast

35

📢 Yet another great dissemination activity of our IPN TechDebtOps project: Two new papers have been accepted for presentation and publication at the 26th International Conference on Product-Focused Software Process Improvement (PROFES 2025), which is among the most recognized conferences in software development and process improvement!  ✒️We wrote the first paper to report our research work on developing DebtGuardian, a tool for “Detecting Technical Debt in Source Code Changes using Large Language Models”. DebtGuardian is an in-house (research-driven) tool developed by our team at SINTEF. We drive it as a complementary tool to the existing commercial tools that we have studied recently. Our multi-tool approaches to Technical Debt (TD) management in industrial C#.NET projects were presented in the ICSME 2025 paper mentioned in another post earlier ➡️ https://lnkd.in/dC_Z99FD More info about DebtGuardian can be found in the post below 👇 🚀Our project partner at UiO got yet another paper accepted at PROFES 2025, which they surely can announce it soon 👏 🌱There is much exciting work and more results to come in the final year of our project. Some are already in the pipeline from the on-going collaboration among Visma, SINTEF, Universitetet i Oslo (UiO) | University of Oslo, AKVA group, and Knowit. The project is funded by Norges forskningsråd and the companies 💚 #LLMs #SourceCode #TechDebt #PROFES #ICSME #SoftwareEngineering #ArchitectureSmells #CodeAnalysis #IndustryCollaboration #SoftwareMaintenance

13

2 Comments

Excited to share our new preprint on arXiv: HOMA — Higher-Order Modular Attention for Protein Sequence Modelling Transformer self-attention is powerful but it only captures pairwise interactions between tokens. For proteins, where function often emerges from the cooperative interplay of three or more residues, this is a fundamental limitation. HOMA addresses this directly by introducing an explicit triadic attention pathway fused with standard pairwise attention through a learned MLP. Rather than replacing pairwise attention, HOMA extends it, giving the model access to both second- and third-order positional dependencies within the same unified operator. To keep triadic attention practical on long sequences, we combine two efficiency techniques: 🔹 Overlapping block structure - shared across both pairwise and triadic branches 🔹 Local windowed interactions - triadic attention is restricted to a sliding window within each block, reducing cost from O(ℓ³) to O(ℓ·w²) per block We evaluate on three TAPE benchmarks: Secondary Structure (SS3), Fluorescence, and Stability. HOMA achieves consistent improvements over standard self-attention, blockwise attention, and Linformer at controllable additional computational cost. This work was carried out under the supervision of Dr. Xin Gao. Thank you for your guidance throughout this project. 📄 Paper: https://lnkd.in/ekXqaJ9h 💻 Code: https://lnkd.in/eC2ys75P Would love to hear thoughts from the community especially anyone working on protein language models or efficient attention mechanisms! #MachineLearning #Proteins #Transformers #Bioinformatics #NLP #DeepLearning #ArXiv #Research

20

1 Comment

📢 New Research Announcement We’re excited to share our latest work from the Distributed & High Performance AI Systems Research Group at the University of Leicester, has been accepted for publication in ACM Transactions on Autonomous and Adaptive Systems, January 2026, ACM (Association for Computing Machinery) DOI: https://lnkd.in/esk_ftQa 🔍 Bandit Neural Architecture Search for SciML Digital Twins introduces a smarter way to automatically design and tune AI models for Digital Twins. Unlike standard AI, physics-based models learn in more structured and predictable ways—and our approach is designed specifically to exploit that. 🚀 By aligning optimisation methods with how scientific models actually behave, this work improves accuracy and efficiency under real-time constraints, helping Digital Twins move from promising prototypes to robust, deployable systems. 🔗 Learn more: https://lnkd.in/eja49hjr

7

2 Comments

The Call for Papers for the second edition of the Bangla Language Processing (BLP) Workshop is out (https://lnkd.in/eT7zRkkF). This year's BLP workshop will be co-located with the IJCNLP-AACL conference from December 20-24, 2025, in Mumbai, India. We invite researchers to submit original work presenting novel approaches and resources that advance Bangla and other low-resource languages. This year's workshop will also consist of two shared tasks (https://lnkd.in/eQGz3v33): Task 1: Hate Speech Detection (https://lnkd.in/eNmX5MCg) Task 2: Code Generation (https://lnkd.in/epkqstEQ) Important Dates: Paper submission due (both regular and shared tasks papers): Sep 29 (Mon), 2025 Notification of acceptance: Nov 3 (Mon), 2025 Camera-ready due: Nov 11 (Tue), 2025 Firoj Alam, Shammur Absar Chowdhury, 🤘 Sudipta Kar, Naeemul Hassan, Enamul Hoque Prince, Md Arid Hasan, Tasnim Mohiuddin, Md Nishat Raihan, Noshin Ulfat

99

We are thrilled to introduce the "Open Orthophosphate Model," an AI and ML-driven framework set to transform the prediction of orthophosphate concentrations in UK water catchments. By utilising millions of historical observations and incorporating advanced molecular embeddings, correlations, and domain expertise, this model provides valuable insights into water quality. This innovative approach aims to aid the water quality monitoring, offering scalable, cost-effective solutions in addressing environmental, regulatory, and technical challenges. Please access the model under the GitHub location: https://lnkd.in/ek6a6QSn and please provide your comments. #WaterQuality #AI #MachineLearning #EnvironmentalTech #Innovation

17

I've become interested in mechanistic interpretability recently and wanted to experiment with some of the latest innovations in the space. In a limited fashion, we can look inside a LLM and see what's going on. Using Sparse Autoencoders (SAEs) from Gemma Scope, this tool lets you peek under the hood of Google's Gemma 2 2B model and actually steer its behavior in real time. You can: - Browse and search 16K+ interpretable features by description - Visualize which features activate on any input text - Amplify or suppress specific features during generation and see how outputs change - Decompose predictions into per-feature logit contributions Built with TransformerLens, SAELens, and Gradio. Try it on Huggingface Spaces: https://lnkd.in/e88v8wym Code here: https://lnkd.in/e7dSS9WA #MechanisticInterpretability #AISafety #MachineLearning #NLP #OpenSource

6

The AI Research Foundations curriculum is a brilliant set of free online AI courses, now available on Google Cloud Skills Boost, with lessons on transformers and generative language models. It is aimed at technical university and community learners (in math, physics, engineering, computer science). I was extremely inspired to see the genesis of this curriculum, built by colleagues including Ulrich Paquet and Avishkar Bhoopchand, the Google DeepMind Impact Accelerator and UCL. You can access it at: https://lnkd.in/eX_EJ6kz

21

The topic of the latent representation of event logs is interesting because it combines aspects related to individual events, inter-case (a sequence of events) and intra-case. Florence Wong has devised an effective strategy that combines semantic completeness and architectural simplicity. GitHub: https://lnkd.in/dY33F3qy arXiv paper: https://lnkd.in/dkpPWzJ2

6