- Transfer Learning in NLP & Text Generation for Fake News Detection
Thanks very much to Deepset GmbH for hosting us, at The Place :):) Talk 1: Transfer Learning in NLP - applied to Question Answering Speakers: Branden Chan & Timo Möller Abstract: Since the transfer learning paradigm came to NLP, models have been able to convert learnings from massive amounts of unlabeled text data into performance gains on downstream tasks like document classification and NER. Another beneficiary of this revolution has been Question Answering, which has seen marked improvements since Google’s BERT model was released. In this talk, we will explain how to adjust a language model to answer questions in automated ways. Since new Language Model architectures are published on a monthly basis, an overview of current models will guide you on how to do state of the art NLP yourself. Bios: Timo Möller is Co-Founder of the Machine Learning startup deepset. He studied Data Science in Maastricht and computational Neuroscience in Berlin, where he also worked several years as ML engineer. Branden Chan is a Stanford graduate in computational linguistics with experience as an NLP engineer - now working for deepset on bringing latest NLP techniques to the industry. Together they have trained large Language Models from scratch and are currently developing a customized Question Answering system for a DAX company. -- Talk 2: Grover vs the fake news robot uprising Speaker: Camille Van Hoffelen Abstract: The debate about the dangers of language generation exploded with OpenAI's refusal to release the full GPT-2 model earlier this year. The conversation is still in full swing today about the best ways to defend ourselves against malicious applications of the technology. A key component of this debate is the ability to automatically differentiate generated language from human language. In this talk, we will first explore state-of-the-art text generation methods using language models. Then, we will breakdown generated text detection techniques, and introduce Grover: our potential saviour from the fake news apocalypse. Bio: Camille is a Research Engineer at Seal Software, where he has spent 5 years building the leading enterprise software in contract analytics. He graduated with an MSci in Physics from Imperial College London. Nowadays, he focuses on implementing state-of-the-art NLP methods for legal AI, with experience in ML infrastructure and software development. He is driven by the impact of cutting-edge AI technologies, and strives to put these big ideas to use in the real world.
- AI for Real-time Visual Analysis
In this meetup, we have two talks about real-time visual analysis. Talk 1: Towards real-time interpretation of the physical world with FPGA and DNNs Speakers: Nicolas von Roden, Hirad Rezaeian Abstract: We aim for a compact, affordable and accurate real-time interpretation engine of the physical world. Processing of high-resolution input data from visual sensors in real-time to achieve correct and high accuracy scene understandings using scalable, robust and price efficient hardware is a huge challenge. We approach this via co-design of FPGA and DNN. On the software side, we exploit multi-task learning to combine different single-task models. On the hardware side, we aim to quantize model weights and activation functions for efficient deployment on FPGA-based hardware. Bios: Nicolas is a Computer Vision Engineer at Advertima AG. He graduated in CS from the U of Erlangen-Nuremberg with a focus on image processing and ML. He is currently working on computer vision tasks for face recognition, pose detection and tracking in real-time as well as combining the various single-task models into a multi-task framework. He previously worked on tumor detection in magnetic resonance images for Siemens Healthineers. Hirad is a Hardware Digital Design Engineer at Advertima AG. He graduated from ETH University as an electrical engineer with a focus on micro electronics and signal processing. His experience in digital signal processing and algorithm improvements regarding hardware implementation (ASIC and FPGA) led him work on ASICs for AI workload acceleration. Currently he is working on the quantization of the model weights and activation functions to reach a ternary weight network on a FPGA-based platform. -- Talk 2: A divide & conquer approach to real time video segmentation on smartphones Speaker: Noah Kutscher Abstract: Real-time video segmentation for extracting humans from images has two challenges: compute speed, and segmentation quality. We are testing a semantic unit to make foreground estimation via a pre-trained Deep Neural Network. This information does not have to be provided in real-time and can therefore have complex computation. This estimation is used in the second stage to train a small but fast model to classify without the need to search for semantic connections between the different image pixels. Both methods are combined for a semantically correct, stable, and fast approach. Bio: Noah is an Undergrad Student at University of Applied Science Mittweida, studying Digital Forensics. After many years of programming, his focus shifted to ML. Since early 2019, he researches real time video segmentation at Cinector.