Machine Learning Course at EPFL, Exercises of Mathematics

Download Machine Learning Course at EPFL and more Exercises Mathematics in PDF only on Docsity! 19th Sept 2023 Machine Learning CS-433 Martin Jaggi & Nicolas Flammarion Master CS-552 Modern Natural Language Processing CS-502 Deep learning in biomedicine EE-559 – Deep Learning (UniGe) CIVIL-459 – Deep learn. for autonomous vehicles EE-411 – Fundamentals of inference and learning MGT-418 – Convex optimization MATH-403 – Low-rank approximation techniques MATH-412 – Statistical ML MATH-520 – Mathematics of machine learning MICRO-455 – Applied ML MICRO-401 – ML Programming MICRO-570 – Advanced ML DH-406 – ML for Digital Humanities CS-430 – Intelligent Agents CS-439 – Optimization for ML CS-401 – Applied Data Analysis CS-503 – Visual intelligence: machines and minds EE-556 – Mathematics of data CS-526 – Learning theory Bachelor CS-233 – Intro to ML CS-330 – Artificial Intelligence EE-311 - Fundamentals of machine learning CIVIL-226 – Introduction to ML for engineers BIO-322 – Intro to ML for bioengineers ME-390 - Foundations of artificial intelligence Seminars, Doctoral Courses and continued education ENG-704 – EECS Seminar: Advanced Topics in ML CS-723 – Topics in ML Systems CS-612 – Topics in Natural Language Processing EE-608 – Deep Learn. for Natural Language Proc. EE-618 – Theory and M. for Reinforcement Learning EE-613 – ML for engineers EPFL Extension School – Applied Data Science: ML Alternatives Research talks: join mailing list: [email protected] on groups.epfl.ch Course Logistics Tuesday 2x45mins, Room: Rolex learning center Thursday 2x45mins, Room: Rolex learning centre We provide PDF lecture notes on our webpage and GitHub, and streaming&recordings of all lectures available Lectures Thursday 14:15 - 16:00 - live interaction! Rooms: INF1, INF119, INJ218, INM202, INR219  assignment by lastname, see course info sheet PDF All labs and projects are in Python. See the first lab to get started. Code Repository for Labs: github.com/epfml/ML_course Course Logistics Exercises Course Logistics Team of assistants Erwan Emlil Vinko Sabolcec Mikhail Seliugin Guanyu Zhang Yauheniya Karelskaya Alejandro Hernandez Cano Naisong Zhou Aybars Yazici contact us: online forum! Lara Orlandic (Organizing TA) Maksym Andriushchenko Francesco D’Angelo Corentin Dumery Dongyang Fan Simin Fan Hojjat Karami Atli Kosson Skander Moalla Hristo Papazov Aditya Varre Oguz Yüksel Philippe Servant Yiyang Feng Leonardo Trentini Simon Halstensen Mathis Randl Mohamed Hadhri Dong Chu Your colleagues here EPFL sections IF ° 2.7% MT_RO 2.3% PhD Bachelor from EPFL ? What is Machine Learning? Introduction What is Machine Learning? learn from data algorithms that can Learning Functions from Data input output hussar monkey image classification “Bonjour! Comment allez-vous?” “Hello! how are you?” translation “Hello! how are you?” speech to text “a dog is sitting at the beach next to another dog” image captioning writing assistant pi xe ls au di o pi xe ls te xt te xt wha…t? input output “moon landing conspiracy” webpage 1 webpage 2 webpage 3 web search browsing history on fashion website recommender system melanoma medical image processing “look at whether it works for the UK or not” lip reading “moon landing conspiracy…” The bot must be trained in a language capable of decoding Python's strings and displaying it on a high quality display, in order to be able to produce what they have learned in English, and indeed, these images, have been uploaded to the web for quite some time. If this type of thing is indeed present in the wild, then what sort of wild bot should I be worried about? Thanks to this one specific experiment performed on the same day - as described by the author: Using some kind of neural network to learn speech, and being able to decode it in order to communicate with others (including yourself) through its GUI text generation professional dancer + photo of myself https://youtu.be/PCBTZh41Ris?t=139 dance transfer pi xe ls vi de o image source towards… understanding intelligence ? if-then-else ≠ intelligence towards… understanding intelligence ? Neuroscience / HBPMachine Learning vs What is the difference between Artificial Intelligence, Data Mining, Statistics, Machine Learning? Cycles of popularity === Machine Learning a= | inear Algebra Note Note Sa Jan 1, 2010 May 1, 2014 Sep 1, 2018 SOUrCe why ML? Applications Industry Applications ✤ majority of industries, originally not ‘digital’: ✤ agriculture, NGOs, ‘sharing economy', logistics, delivery, services, manufacturing, sports, personalized health, call centers, entertainment, … ✤ not only the ‘usual suspects’ History ✤ ML is not new! London 1854 cholera outbreak History ✤ ML is not new! ✤ the early days - 1950ies and 1960ies ✤ Neural networks ✤ Turing What has changed? 1950s: 103 FLOP 2023: 1024 FLOP* “the embryo of an electronic computer that … will be able to walk, talk, see, write, reproduce itself and be conscious of its existence.” 1958 *chatGPT training estimate Challenges ✤ Hype ✤ cycles of AI popularity ✤ Data Ethics, Privacy, Fairness ✤ Lack of Interpretability ✤ example: medical applications of deep learning ✤ Social Implications of AI ✤ impact and potential unsafe use (today’s models) ✤ e.g. impact on labor market, education, abuse, misinformation,… ✤ threats from Super-human AI (tomorrow’s models) ✤ see Nick Bostrom, Yuval Harari need: Scientific Method, Reproducible Research, Open Source and Open Data ML Applications by CS-433 Master Students Machine Learning-based Estimation of Cardiac Contractility from Peripheral Pressure Waveform Deep learning techniques for geometric matching of C. Elegans brain microscopy images Benchmarking Machine Learning Methods for Eukaryote/Prokaryote Contigs Classification Machine Learning for Science: Classification of Skin Samples Using Mass Spectrometry Analysis Application of Deep Knockoffs for fMRI to Generate Surrogate Data Automatic detection of weak cipher usage in aircraft communications Predicting Topic Change and Emoji Usage from Twitter Data Cell Nuclei Segmentation in 2D Fluorescence Microscopy Images Unsupervised cell classification in flow cytometry data Predicting chemicals concentration in water streams using Gradient Boosting Regressor Extracting Masonry Building Facades through Polygon Image Segmentation Sequence-dependent clustering of DNA in Protein-DNA Xray crystal data and in cgDNA+ model Applying the VoxelMorph Framework to C.Elegans Brain Data Using forearm sEMG to control individual fingers of a robotic hand Music beyond Major and Minor Avalanche Forecasting: An Ordinal Regression Approach Machine Learning for Side-Channel Disassembly Multi-object Detection and Tracking Motion-based Similarity Search in Videos of Confucian Rituals Detecting the Degree of Cavitation In Situ in Young Trees Machine Learning in Chemistry Personalized Federated Image classification using Weight Erosion In-crystal Gamma-interaction localization for positron emission tomography (PET) from Cherenkov photons Classification of zebrafish embryo using various ML methods Resource-Efficient Machine Learning Algorithm Design for On-Implant Neurological Symptom Detection Ebola Virus Disease Diagnosis for West African Ebola Virus epidemic Supervised classification of fly behaviors from posetracking data Cell-type classification from microscope imaging COVID-19 Predictions using Machine Learning Unsupervised time series analysis of country wise COVID data Voxelmorph Unsupervised classification of video games styles Can the Style and Wording in Critical Reviews of Video Games Predict its PEGI Labelling? Ensemble Methods for Dynamic Portfolio Valuation Vector embeddings of harmonies in music with deep learning Robustness of U-Net based models to common image artefacts Recognizing Humor and Predicting Humor Ratings in Short Texts Segmentation of cell nuclei in 2D microscopy images with CNNs Mechanism of Action (MoA) Prediction – Kaggle Competition Diagnostic and Prognostic models for Ebola Automatic Grading of Handwritten Student Essays Stroke Level Prediction through Pacman Game Data Among Us Project 2 – Market states prediction Regularized maximum likelihood estimation – TRANSP-OR Stroke Level Estimation through pac-man game data played by acute stroke patients STLM: Steganography in Text using Language Models Eastern Rituals Search Engine (ERSE) Cough Classifier Extracting high value lung ultrasound images from video for the diagnosis and prognosis of COVID-19 Detecting rooftop available surface for installing PV modules in aerial images using Deep Learning Dimensionality reduction and clustering of energy consumption time series in supermarket buildings Protein-Protein Interactions Predicting gene-gene relationship with CNNC model PneumoNet: Neural networks for the detection of pneumonia from digital lung auscultation audio Predicting errors during Pacman for stroke patients Galaxy Detection Machine Learning Project Automatic detection of available area for rooftop solar panel installations Prediction of myocardial infection risk after stenosis diagnosis LC3 compressive strength analysis Adapting Attention Guided Camera Localization for the Geodetic Engineering Laboratory Machine learning models to predict the diagnosis and risk of COVID-19 from clinical data in Switzerland Facades and Openings Detection Based on Different Deep Learning Models Variational Inference compared to Markov Chain Monte Carlo for modelling gene expression 3D Spatiotemporal clustering of mixed-type medical data in Tanzania Classification and Clustering on Schizophrenic Patient’s Data TRANSP-OR – Prediction of mode of transportation Learned cross-domain descriptors (LCD) for drone navigation What if Interactive GlobalCOVID Policy Simulator Image Segmentation of Adenovirus Particles in Food Vacuoles of Eukaryotic Organisms Music Beyond Major and Minor Determining the important features for estimating the reproduction number in the COVID-19 pandemic Exploring chord embedding spaces between musical composers and eras Vector Embeddings of Musical Chords Word embeddings and transformer models for optimal learning Identification of fire periods from air quality monitoring network measurements Drone and pigeon detection Characterization of turbulent structures in tokamaks Improving Deep Learning models for EMG decoding used for prosthesis control enhancement Pneumonia Diagnosis based on CNN-LSTM-BERT Model L-form bacteria segmentation Machine Learning for Spaced Repetition in Human Learning COVID-19 risk stratification on Chest X-Rays: performance on a small cohort of patients in Switzerland Dry vs Wet Cough Automatic Classification using the COUGHVID Dataset Improving Freshwater Quality Measurements through Machine Learning Lesion detection on cardiology images using Deep Learning 3D to 2D feature matching for next generation 3D mapping algorithms Calibrate a model of OTC markets ML4science projects by your colleagues 2020 Reproducibility Challenge Reproducibility Challenge recent examples: On Warm-Starting Neural Network Training Can gradient clipping mitigate label noise? Sanity-Checking Pruning Methods: Random Tickets can win the Jackpot Learning to Play Sequential Games versus Unknown Opponents Distributed Distillation for On-Device Learning AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer Reproducing Empirical Results and Assessing Theoretical Guarantees of PAGE Contrastive Learning of Musical Representation Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalization Learning by Turning: Neural Architecture Aware Optimisation reproduce (parts of) the experiments in a machine learning research paper: https://paperswithcode.com/rc2022 (To be confirmed for 2023) (the papers can be from any conference such as 
 NeurIPS, ICML, ICLR, ACL-IJCNLP, EMNLP, CVPR, ICCV, AAAI and IJCAI Applications, cont. why ML? Image Data ✤ Astronomy ✤ Face recognition ✤ 2D + 3D medical imaging ✤ OCR ✤ self-driving cars how-old.net (deactivated) Audio & Multimodal Data ✤ Hearing aids ✤ Voice Recognition ✤ Automatic Translation ✤ Lip Reading ✤ Video Analysis Numerical / Sensor Data ✤ Cern ✤ Astronomy/ Telescopes ✤ Fitness Trackers ✤ Weather Forecast ✤ Robotics ✤ Kinect Games & Simulations ✤ Immediate Feedback ✤ Chess, Go ✤ Physical World (Generative generation of image, video, audio, text modalities ur World Timeline of images generated by artificial intelligence Su These people don’t exist. All images were generated by artificial intelligence. 2014 2015 2016 ed Radford, Metz, and Chintala (2015) - Unsupervised Liu and Tuzel (2016) - Coupled GANs Goodfellow et al. (2014) - Generative Adversarial Networks Representation Learning with Deep Convolutional GANs ~ 2017 Karras, Laine, and Aila (2018) - A Style-Based Generator Karras et al. (2019) - Analyzing and Improving Karras et al. (2017) - Progressive Growing of GANs Architecture for Generative Adversarial Networks the Image Quality of StyleGAN for Improved Quality, Stability, and Variation : 202 O 20 2 Image generated with the prompt: 2 O 2 Image generated with the prompt: a couple of people are sitting on a “A Pomeranian is sitting on the King’s wood bench throne wearing a crown. Two tiger soldiers are standing next to the throne.” SS - Zero-Shot Text-to-Image Generation Saharia et al. (2022) - Photorealistic Text-to-Image Diffusion Ramesh et al. (2021) Models with Deep Language Understanding (Google's Imagen) Ho, Jain, & Abbeel (2020) - Denoising Diffusion Probabilistic (OpenAl's DALL-E 1) Models OurWorldinData.org - Research and data to make progress against the world’s largest problems. Licensed under CC-BY by the authors Charlie Giattino and Max Roser New Opportunities? Your turn up next: ✤ Regression ✤ Linear Regression ✤ Classification ✤ … fundamental concepts of ML