Nick Talken’s Post

Delighted to share a new publication, "Spectral Filter Predictor for Progressive Latent Fingerprint Restoration." This breakthrough method restores latent fingerprints using deep learning to predict frequency-domain filters instead of working directly in the spatial domain. Integrated into a progressive feedback framework, it reveals genuine friction ridges from low-quality prints with complex backgrounds. Benchmarking against four public datasets, this method outperformed current state-of-the-art techniques, improving rank-1 identification accuracy by 3.75% with a commercial matcher and 3.19% with an open-source matcher. This marks a significant advancement in latent fingerprint identification! Paper: https://lnkd.in/gRa63zzU GitHub: https://lnkd.in/gHHygBtV

Today, we shine a light on Laeticia Onyejegbu, a young researcher who attended the 9th HLF. Nowadays, Laeticia is a Professor of Computer Science, Artificial Intelligence and Machine Learning. This article provides thrilling insights into her passion for working with AI and the potential she sees in using AI to solve big societal problems. Get to know more about her inspiring research here: https://lnkd.in/gQZkhwrg

We are proud to share this new case study describing the measurement of ground cover in grazed destocked hill and high-country farmland across Canterbury using Machine Learning. It was great working alongside experts at Environment Canterbury including Jeromy Cuff, Zach Hill and Laurence Smith to automate the classification and measurement of this important and continuous data set which which began in the 1970s. https://lnkd.in/gV-FDFKR Lynker Analytics was commissioned to develop a convolutional neural network (CNN) semantic segmentation model to automate the classification of each photo into its fractional ground cover to enable scientific interpretation.  Led by Principal Data Science David Knox the models developed during this research allowed for automated inference across the whole photo archive and consistent classifications across time. This work was also published at the 38th International Conference on Image and Vision Computing New Zealand (IVCNZ). The examples shown here include: observed image, training points, and inference . Red = Dead, Green = Living, White = Rock, Blue = Soil.

Excited to share that a paper describing our latest method, ELLA, has been accepted at the #neurips2023 conference in New Orleans (Awarded with spotlight). The paper deals with the question whether layer-wise equivariance symmetries in machine learning algorithms can be learned automatically from data. Equivariance symmetries (e.g. conv layers) allow neural networks to generalise effectively, but are typically fixed, need to be chosen in advance, and can not be adapted. We show that we can learn them automatically using gradients! A summary of the paper can be found in this X/Twitter thread: https://lnkd.in/d8VKzmes An open-access pre-print of the paper is available on Arxiv: https://lnkd.in/dUxN5eku. Code is available on Github: https://lnkd.in/dV3rHJ27. Feel free to reach out with a pm/email if you're interested! 🙏 Great collaboration with Alexander Immer and Mark van der Wilk.

🚨 What does it take to integrate AI into real-world healthcare practices? Welcome to the 13th edition of the Neural Medwork Newsletter https://lnkd.in/e4FRydbE! This time, we're shifting our focus from the Perceptron to explore Support Vector Machines (SVM) in machine learning. 🔍 Dive deep our latest research paper that sheds light on the practical implementation of AI solutions in healthcare settings. 🧠 We're also excited to introduce the Advanced Reasoning with Tools (ART) technique, a game-changer for employing large language models (LLMs) in tackling complex healthcare challenges.

Hope to see you all there. This should be a great conversation about Generative AI in education--with thinking about Gemini, ChatGPT, and other tools--and what they may mean for education.

🌟 𝗙𝗿𝗼𝗺 𝗕𝗲𝗿𝗹𝗶𝗻 𝘁𝗼 𝗛𝗮𝗿𝘃𝗮𝗿𝗱: 𝗠𝗲𝗲𝘁 Jonathan Geuter - 𝗢𝘂𝗿 𝗠𝗮𝘁𝗵𝗲𝗺𝗮𝘁𝗶𝗰𝗮𝗹 𝗠𝗮𝘃𝗲𝗿𝗶𝗰𝗸! 🌟 Ever wondered what happens when raw academic talent meets groundbreaking AI innovation? Dive into Jonathan's journey with Jina AI! A prodigy from Technische Universität Berlin with a Bachelor's and Master's in Mathematics, and now making waves at Harvard University with his Ph.D. in Applied Mathematics, zeroing in on machine learning. But that's not all! 🚀 During his stint at Jina AI, Jonathan didn't just learn. He led. He tackled the challenge of fine-tuning embedding models using LLMs and smashed through the State-of-the-Art baselines. Curious about his findings? Check out our latest paper uploaded on arXiv! https://lnkd.in/gAveyq9p 🎥 𝗪𝗮𝘁𝗰𝗵 𝗵𝗶𝘀 𝘀𝘁𝗼𝗿𝘆 𝘂𝗻𝗳𝗼𝗹𝗱 𝗯𝗲𝗹𝗼𝘄 Like and Share if you believe in the power of academia and industry coming together! 🔥🔥 Let’s inspire the next generation of multimodal AI visionaries.

Hello All, A recent work from our lab on Neural discovery of Discrete Symmetries got accepted in AISTATS 24 (A premier conference for AI theory). In this work, We consider the problem of learning a function respecting a symmetry from among a class of symmetries. We develop a unified framework that enables symmetry discovery across a broad range of subgroups including locally symmetric, dihedral and cyclic subgroups. This is a rare paper where we have 4 reviews with a unanimous accept even in the pre-rebuttal phase. Congratulations to my PhD student Pavan Karjol, RA Rohan Kashyap and colleague/collaborator Aditya Gopalan. Here is the Arxiv version for interested readers - https://lnkd.in/gwBMbgEG

AI Starts to Sift Through String Theory’s Near-Endless Possibilities. Using machine learning, string theorists are finally showing how microscopic configurations of extra dimensions translate into sets of elementary particles — though not yet those of our universe - https://lnkd.in/gTaUaVRH

Love this reality check - AI is a tool, understand its strengths & weaknesses & use for what is it good for….