Hiking through the wilderness of neural network loss landscapes – Dr Anna Bosman Abstract Deep neural network training is a highly non-convex optimisation problem with poorly understood properties. We know that a solution can be found by following the negative...

An Introduction to Variational Inference and its Application in Deep Learning – Jacobie Mouton Abstract Bayesian inference allows us to calculate the posterior distribution of unknown variables given observations, using Bayes’ Theorem. In practice however, it is typically the case...

Shocklab x InstaDeep x UCT AI Society: Exclusive Film Screening SPEAKER Presented by InstaDeep and AI Society DATE 19 September 2023

Beyond Python: Why you should consider Julia for your next reinforcement learning project – Sasha Abramowitz Abstract This talk covers a brief intro to Julia programming language. It then compares it to the other options out there for reinforcement learning...

Voice conversion with just nearest neighbours – Matthew Baas Abstract Voice conversion aims to transform speech into a target voice with just a few example recordings of the target speaker. Recent methods produce convincing conversions, but at the cost of...

Partially Automating the Improvement of Learning Agents (PAILA) Abstract The PAILA project, undertaken during our InstaDeep internship, aims to bolster single-environment Reinforcement Learning (RL) algorithms through cross-environment knowledge sharing. To achieve this, we aimed to use symmetric learning agents (SymLA),...

Denoising Diffusion Models: Introduction and Applications Abstract Denoising Diffusion Models are a type of generative modelling which serves backbone of recent advances in image synthesis including Dall-E 2, Midjourney, and Imagen. These models utilise an iterative denoising process during inference...

Modular Evolutionary Origami Robotics Abstract Evolutionary robotics lends itself to exploring novel design paradigms in research to assess the efficacy of those designs relative to known paradigms in the space. Origami is one such paradigm that has been relatively under-explored,...

Surveying research directions on AI safety Abstract AI safety is a subject which has often been viewed with skepticism regarding its necessity and plausibility in the AI community. However, as we have progressed towards transformational AI systems the urgency of...

Efficient Inverse RL Abstract Interactive learning systems like self-driving cars, recommender systems, and large language model chatbots are becoming increasingly ubiquitous in everyday life. From a machine learning perspective, the key technical challenge underlying such systems is that rather than...

The Impact of Morphological Diversity in Robot Swarms Abstract In nature, morphological diversity enhances functional diversity, however, there is little swarm (collective) robotics research on the impact of morphological and behavioral (body-brain) diversity that emerges in response to changing environments....

Molecule Design Based on Multi-objective Optimisation and Graph Transformers Abstract I will be presenting an empirical exploration of using machine learning and evolutionary algorithms to automate chemical product design. Our study demonstrates how computational design can be controlled via hyper-parameters...

Simulating the Past, Present and Future Using Agent-Based Models Abstract Humans are fundamentally social creatures, we live in families, work in teams and our norms of formed from thousands of years of social interaction. What if, along that timeline, something...

Intuitive explanations of the transformer model Abstract In this talk I want to explain in as clear a way as possible what the key concepts are in a transformer model, explain key terms, and discuss why the transformer is so...

Supporting RL Evaluation with Multi-Criteria Decision Analysis Abstract The evaluation of empirical algorithm performances in RL appears a closed topic. However, some (sparse) recent research provides unattended criticisms of key elements of the evaluations which are central to the conclusions...

AI 4 Health in Production – Africa Abstract I explore the challenges facing production AI for health systems in an African context. Progressively I step through the layers of complexity, one can expect to encounter, providing personal insight for addressing...

A Folk Theorem from Learning in Games Abstract We introduce a generalisation of smooth fictitious play with bounded m-memory strategies. We use this learning algorithm to prove a Folk theorem from learning in repeated potential games. If a payoff profile...

Selective Reincarnation in Multi-Agent Reinforcement Learning Abstract Claude presents his work on selective reincarnation for MARL. SPEAKER Claude Formanek DATE 5 April 2023

PyTorch and Weights and Biases for ML Abstract Jeremy give’s an overview of PyTorch and Weights and Biases, emphasising how these are useful for ML in production and in research. SPEAKER Jeremy du Plessis DATE 22 March 2023

NeurIPS in a nutshell Abstract Ruan’s highlights and takeaways of NeurIPS 2022. SPEAKER Ruan de Kock DATE 15 February 2023

Visual cortex is optimised for short timescale prediction using spikes Abstract A key question in systems neuroscience is to understand what principles underly the sensory processing throughout the brain. Why are certain neurons in V1 selectively tuned to orientated bars?...

Towards Lifelong Reinforcement Learning through Logical Skill Composition Abstract Reinforcement learning has achieved recent success in a number of difficult, high-dimensional environments. However, these methods generally require millions of samples from the environment to learn optimal behaviours, limiting their real-world...

Generalisation in ML Abstract In Reinforcement Learning there is often a need for greater sample efficiency when learning an optimal policy, whether due to the complexity of the problem or the difficulty in obtaining data. One family of approaches to...

Offline MARL and how to use WANDB effectivly for ML experiments Abstract Claude gave a talk on his research topic, Offline MARL, and also gave a tutorial on how to use Weights and Biases for ML experiments. SPEAKER Claude Formanek...

Generalisation in a Nutshell Abstract Ruan presents an overview of generalisation in RL. SPEAKER Ruan de Kock DATE 21 September 2022