Deep Learning Workshop : Applied Computer Vision

DESCRIPTION   Deep Learning Workshop : Applied Computer Vision  This practical session will focus on using deep learning for applied computer vision tasks. It will focus on the different families of models (CNN, ViT), how they work and how to train them. Both theory and practical implementations will be covered. After this course, participants will have the basic knowledge to train and implement their own computer vision models, as well as understand best practices for training, monitoring experiments, fine  and selecting models. TARGETED AUDIENCE Industry professionals looking to learn more about applied computer vision and deep learning Academics Students (Bachelors, MSc. PhD) with an interest in deep learning Enthusiasts curious to learn about deep learning and computer vision LEARNING INTENTIONS Introduce professionals to new technologies in machine learning, that are most relevant in the financial sector Develop an understanding of the challenges and issues of data science applied to investments, banking and insurance Learn to use computer tools to solve concrete problems SESSIONS   Session 1 - theory Duration: 2 hours Content: Understanding what images are and how computers interpret them (grayscale vs. RGB images) Pixels Tensor, matrix, vector representations How to sanitize inputs to a model (resized images, cropping, dims, mean subtraction, etc.) image normalization (mean subtraction), channel swapping Explain the concept of a batch Do a demo code demo with mario, with PIL, numpy and torch Explain the basics behind image classification in a supervised learning setting Session 2 - practical Duration: 2 hours Summary: This will be a hands-on, live-coding session. The task will be to implement a rock, paper, scissors game using vanilla pytorch. Models must be vanilla pytorch, relatively simple models (e.g. LeNet). 5 epochs maximum for training. Images should be squared, and resized.  Session 3 - theory Duration: 2 hours Type: theory Content: Understanding the basics of the main families of models in supervised learning (MLP, CNN, ViT) Pros, cons Intuitions Operations definitions Tradeoffs, benchmarks, speed, inference, model size etc. Dataset sizes How to fine tune models vs. training from scratch data augmentation techniques Foundation models + zero shot learning (CLIP) [optional, time dependent] Do a comparison of CLIP zero shot vs. pretrained + fine tuned model? Session 4 - practical  Duration: 2 hours Type: practical Summary: In this session, not much coding is actually happening. What we will be doing is looking at an implemented experiment (all code on github) where we train on a larger dataset (e.g. each trial can take hours). The task should remain around basic classification, maybe do some multi-label classification. However, we look at best practices surrounding the experiment - monitoring, testing, reporting, checkpointing etc. and how all these things can be implemented. TRAINERS Jeremy Pinto    Applied Research Scientist LinkedIn     WORKSHOP TRAINING Duration : 8 hours - over 2 weeks - ONLINE   Level: Intermediate to advanced  Language:  English October 4th -14th 2022 4 sessions - e very Tuesday and Friday 9:00 AM to 11:00 AM (HAE, UTC-4) TRAINING CERTIFIED BY SCALE AI 50% discount on General Admission for professionals  Canadians Promo Code: IVA_Scale_DLW0922 LEARNING OBJECTIVES Apply deep learning to image classification tasks Prepare and process images to pass them through deep learning models  Examine/Compare the architectures and characteristics of the main deep learning vision algorithms (e.g., Convolutional Neural Networks, Vision Transformers) and pretrained models (e.g. ResNets, ViTs)  Train a model using the Implement best practices(e.g., experiment management, code versioning, model checkpointing, etc.) Properly evaluate the performance of the models (e.g., accuracy, f1 score, confusion matrix, etc.) PREREQUISTE Basic Linear algebra concepts (matrix multiplication, vectors, dot products, etc.)  Background in Machine Learning is recommended but not strictly necessary (supervised learning, gradient descent + back propagation)  Basic programming (working knowledge of Python or similar language) is strongly recommended for the tutorials PROGRAM Schedule October 2022    Basics of image proc and data prep Tuesday Oct. 4th  - Session 1: Theory Friday. Oct.7th   - Session 2: Practical Training and evolution of DL model Tuesday Oct. 11th - Session 3: Theory Friday Oct. 14th - Session 4: Practical REGISTRATION FEES*** General Admission  600$ + tx** Student*  100$ + tx** *On presentation of a university ID. (Send it to ) ** Prices are in Canadian dollars, and local taxes (GST, QST) are only applicable for Canadian participants. *** SCALE AI Discount is only applicable to Canadian citizens and / or people working in Canada. For other participants, general admission applies. Registrations end on Friday September 30th, 2022 ADDITIONAL INFORMATION Attestation For the delivery of a certificate for a successful completion to this online workshop, participants must attend the live sessions on a weekly basis. Note to IVADO members  If as an industrial member you have received a member promo code, and, as you are also eligible to Scale AI as a working individual, you may combine both promo codes upon general admission registration.  Check with your manager for promotional codes before registering.  Please note that promo codes are non-transferable and non-exchangeable ASSISTANCE AND SUPPORT For information on programming: