Prof. Yoshua Bengio

A.M. Turing Award

  

Cognitively-inspired inductive biases for higher-level cognition and systematic generalization

Humans are very good at out-of-distribution generalization (at least compared to current AI systems) and it would be good to understand some of the inductive biases they may exploit and test these theories by evaluating how they can be translated into successful ML architectures, training frameworks and experiments. Natural language and experimental results in cognitive science and neuroscience provide a wealth of clues about the system 2 part of how humans understand the world and consciously reason about it. In this talk, I will discuss several of these hypothesized inductive biases, many of which exploit notions in causality and connect the discovery of abstractions in representation learning (the perception and interpretation part) and in reinforcement learning (the abstract actions). Systematic generalization is hypothesized to arise from an efficient factorization of knowledge into recomposable pieces corresponding to reusable factors. This is related yet different in many ways from symbolic AI (and this can be seen in the errors and limitations of reasoning in humans, as well as in our ability to learn to do this at scale, with distributed representations and efficient search). Sparsity of the causal graph and locality of interventions -- which can be observed in the structure of sentences -- have the potential to considerably reduce the computational complexity of both inference (including planning) and learning, which may be a reason for which evolution may have incorporated this "consciousness" prior. Although this talk will rest on a series of recent papers on these topics (e.g., on learning causal and/or modular structure with deep learning), much of it will be forward-facing and suggest open research questions in the hope of stimulating novel investigations and collaborations

Prof. Ali Idri

  

Data Preprocessing in Medical Knowledge Discovery

Current information and storage technologies are resulting in the explosive growth of many business, government, and scientific databases. This has led to the development of advanced techniques and tools to assist humans extract useful information and make informed decisions from available data. Knowledge discovery in databases (KDD) has therefore become a very active research field and its applications may range from business management, and market analysis, to engineering design and medical exploration. KDD is concerned with the development of powerful and versatile tools for making sense of data. It consists of three main steps: data preprocessing (DP), data processing or data mining (DM), and knowledge evaluation and validation. DM is the mathematical core of KDD which deals with the application of intelligent techniques in order to obtain useful patterns, while DP deals with different real-world data imperfections for a successful use of DM techniques. In medicine, KDD can be used to extract knowledge from clinical data for effective medical diagnosis, prognosis, treatment, screening, monitoring, and management. However, medical data are usually incomplete, imbalanced, and contain outliers, noise, and inconsistencies along with redundant and irrelevant variables. This makes DP a necessary step before any DM algorithm is applied. In fact, about 30-70% of effort and time is usually spent on DP, while the DM step requires only about 10-25% of the total effort of a KDD project. Applying DP to medical datasets can have several benefits such as improving clinical interpretability, reducing the computational cost, improving the performance of the predictors and reducing their complexity.

Prof. Hajar Moussannif

     

Artificial Intelligence: Unlocking Real Value Beyond the Hype.

From Virtual Assistants communicating with humans and offering relevant responses to their queries in real time, to data analytics tools amassing torrents of data and using deep learning to sift through them and recognize patterns, the Artificial Intelligence field is growing at an unprecedented pace. AI is reshaping the way humans think, operate and live. The aim of this talk is to highlight the trends, as well as the emerging areas in the AI field beyond the hype, covering interesting applications from both research and business perspectives, with special focus on generative modeling, transformer and tinyML.

Dr. Halima Bensmail

     

Harnessing big 'Omics' data with ML and AI for disease signature discovery

Machine learning is becoming popular and frequently used to solve complex problems particularly in biomedical applications. In this talk we present and analyze three areas in which ML and AI have been used to dissect valuable information. Those areas are related to Multiple tissue transcriptomics, RNA-Seq rare genetic disorders discovery and protein crystallization.

Kemal El Moujahid

  

Google Machine Learning Ecosystem for Research and Production

Kemal El Moujahid will introduce the Google Machine Learning ecosystem, and how it bridges the gap between Research and Production. Kemal will share how TensorFlow, JAX, TPUs, help researchers push the boundaries of ML, and empower developers to solve problems they are passionate about.

Dr. Kaoutar El Maghraoui

     

Powering the Future of AI through Specialized AI hardware Accelerators

AI and especially deep learning have achieved incredible performances in numerous fields including Computer Vision, Speech Recognition, Natural Language Processing etc. However, with this huge success, comes the increased size and complexity of AI models which both translate into high computational and increased carbon footprint. To address these challenges, there is Cambrian explosion of innovative AI hardware-accelerator architectures optimized for deep learning and machine learning across cloud and edge platforms. Purpose-built hardware will shift the traditional balances between cloud and edge, structured and unstructured data, and training and inference. This talk uncovers the evolving landscape of specialized AI hardware accelerators. It also highlights IBM Research suite of techniques towards the design & build of optimized deep learning hardware as part of IBM’s AI hardware Center initiative. We are making further strides in AI hardware-software co-design using approximate computing principals and Analog non-Von-Neumann approaches to unlock exponential gains of AI computations making AI faster, more efficient, and sustainable.

Mehdi Ghissassi

  

Building Research grounded products that impact billions of users

I will be sharing lessons from the past 5 years partnering with Google teams to build products that are grounded in the research and scientific breakthroughs coming out of DeepMind's research teams.

Prof. Mohammed Essaidi

     

Morocco's Strategies and Policies to Leverage AI For Its Socioeconomic Development

Artificial Intelligence presents huge opportunities and an unprecedented potential for socioeconomic and sustainable development through the amazing tools it provides such as Machine / Deep Learning, that allow to get insight from Data, and processes automation and optimization in all kind of areas such as industry, education, healthcare, agriculture, energy, etc. According to McKinsey predictions, AI may deliver an additional economic output of around US$13 trillion by 2030, increasing global GDP by about 1.2 % annually. Most countries around the world are establishing national AI strategies and policies to tap into these amazing opportunities. During the last few years, Morocco has put in place several strategies and public policies to tap into the outstanding opportunities of AI in key areas as diverse as industry, education, public services, healthcare, agriculture, smart cities, etc. This keynote talk will give an overview of these strategies and policies and will also focus on all kinds of indicators related to Morocco's AI readiness and its impact on different areas.

Ulrich Paquet

     

AlphaZero: A Retrospective Story

As a chess-playing teenager in South Africa, I followed the match between Garry Kasparov and IBM's Deep Blue from a long distance. It was an exciting time and an exciting moment in the history of AI, computing and chess. As a youngster that was still only coding in Pascal, I wished that I had an insider's view of what IBM's system looked like. Fast forward more than two decades, and there was another exciting moment in the history of AI, computing and chess. DeepMind's reinforcement-learned system called AlphaZero entered the scene. It learned to play jaw-dropping chess itself, either ignoring conventional human wisdom or showing deep human-like instinct. This talk will give an insider's view of AlphaZero: what it is, how it is trained, what its impact has been, and what AI research has blossomed around it.

Walid Daou

  

How AI can Create Value for Corporates : History and Use Cases

Even if more and more companies are building Data Science teams internally, a lot of AI projects do not create as much value as expected; why so? And how concretely Data Scientists can create value for corporates, especially for non tech companies?