Maties Machine Learning https://mml-stellenbosch.github.io/ Meet - - PowerPoint PPT Presentation

Maties Machine Learning

https://mml-stellenbosch.github.io/

Meet the MML Research Groups

• Arina Britz
• Bruce Watson
• Corné van Daalen
• Hugo Touchette
• Sugnet Lubbe
• Trienko Grobler

cair

Knowledge Abstraction, Representation and Reasoning in Artificial Intelligence

Arina Britz

Centre for AI Research, Dept of Information Science, Stellenbosch Univ, South Africa abritz@sun.ac.za

2019

Cognitive Computational Logics

Logics for Cognition and AI:

◮ description logics; ◮ modal logics

Knowledge Abstraction and Representation:

◮ language design; ◮ expressivity; ◮ applications

Reasoning:

◮ entailment; explanation; learning; debugging; interoperability; ...

cair

Future Perspectives

Reasoning support:

◮ extending tools with non-classical reasoning capabilities; ◮ knowledge abstraction from data

Methodological support:

◮ ontology debugging, revision, repair; ◮ domain visualisation and exploration

Integration of KR with other formalisms:

◮ Integration with formal concept lattices; ◮ Integration with machine learning; ◮ Explainable AI

cair

cair

Applied Algorithmics

Bruce W. Watson

Information Science Chairman, Centre for AI Research Centre for Knowledge-Dynamics & Decision-Making bwwatson@sun.ac.za

2019

Algorithmics

Correctness-by-construction (CbC)

◮ Calculus for algorithms and programs, correctness proof ◮ Used in conjunction with verification and testing ◮ Extend into X-by-C and parallelism

Inventive algorithmics. . . come up with entirely new ones

◮ Use CbC to invent entirely new algorithms ◮ Stringology ◮ Finite automata ◮ Glass box knowledge representation (lattices)

Domain-specific implementation techniques

◮ Finite state techniques ◮ IoT, FPGA, GPU

cair

AI & Algorithmics for Cybersecurity

Pattern correlation

◮ Network traffic learning and scanning ◮ Distributed pattern correlation ◮ Complex Event Processing

Generation of cyberweaponry

◮ Machine learning for hybridizing

Decision- and supply-chains

◮ Group decision support & vulnerabilities ◮ Signing of the silicon IP supply chain

Cryptography

◮ CbC for post-quantum cryptography

cair

People with the department

A mix of internal/extern, all with Ph.D’s and mostly professors Loek Cleophas (Eindhoven) — computing scientist, engineer Fritz Solms (S-plane) — physicist, software engineer Tinus Strauss (Pretoria) — computer scientist, gentleman scientist Derrick Kourie (Pretoria) — computer scientist, OR/stats Norbert Gronau (Potsdam) — computer engineer, i4.0, head of institute Martin Berglund (Munich) — computer scientist, mathematician Jackie Daykin (KCL, Wales) — mathematician, met Erdos Bruce W. Watson — computing scientist/engineer, discrete mathematician, chip nerd

cair

Probabilistic Reasoning and Planning for Mobile Robots

Probabilistic Reasoning and Planning for Mobile Robots

Corn´ e van Daalen

Electronic Systems Laboratory (ESL) Department of Electrical and Electronic Engineering http://staff.ee.sun.ac.za/cvdaalen/ cvdaalen@sun.ac.za

1 March 2019

Probabilistic Reasoning and Planning for Mobile Robots

Applications

Probabilistic Reasoning and Planning for Mobile Robots

Overview and Approach

Sensors: stereo cameras, lidar, radar Problem: autonomous navigation

mapping, localisation, planning large, online, uncertain

Approach:

probabilistic modelling inference and planning under uncertainty probabilistic graphical models (PGMs)

Probabilistic Reasoning and Planning for Mobile Robots

Projects

Mapping: 3D mapping with uncertain robot pose Detection of moving objects using stereo vision Multi-object tracking with radar sensors Modelling of and inference in a semantic map Localisation: SLAM with limited resources Semantic SLAM for drone localisation

Probabilistic Reasoning and Planning for Mobile Robots

Projects

Planning: Robot planning under uncertainty Probabilistic collision prediction Probabilistic reasoning (general): Robust inference

Research on stochastic processes

Hugo Touchette

Applied Mathematics Stellenbosch University

x(t) t AT

• Noisy systems (Markov process)
• Rare events (fluctuations)
• Rare transitions (jumps, crashes, etc.)
• Simulations and sampling
• Prediction, estimation

Hugo Touchette (Stellenbosch) Stochastic processes 1 / 2

Current projects

P(AT = a) ≈ e−TI(a)

Large deviations

• Compute I(a) for specific models
• Markov processes, spectral methods

Rare event simulations

• Efficient sampling of P(AT = a)
• Monte Carlo, importance sampling

Change point detection

• Find locations where statistics change
• ML prediction, sequence modelling
• Students:
• Johan Du Buisson MSc Phy
• Stuart Reid

MSc AM

• Wessel Blomerus

MSc AM

• Faith Msibi

Hons AM

• Post-doc:
• Daniel Nickelsen

SU Phy

• Collaborators:
• Raphael Chetrite

U Nice

• Arnaud Guyader

U Paris

• Grant Rotskoff

NYU

Hugo Touchette (Stellenbosch) Stochastic processes 2 / 2

Multivariate visualisation put to practice

Multi- dimensional scaling MDS

 Visualisation of multivariate data based on some dissimilarity  Biplots

 Subset of MDS  Simultaneous display of rows and columns of a data matrix  Simplest example: PCA biplot  Challenges: big data  Long data – represent cloud of points with 𝛽-bags  Wide data - ???

Projects

 Niël le Roux & Sugnet Lubbe

 Book on Canonical Analysis with John Gower (UK)  Visualization of class separation in the two-group case  Computing and validating neural reliability measures derived from EEG recordings with Pieter Schoonees (Erasmus University)  High throughput sequencing data  Compositional data  Clustering / Classification  Functional data analysis

Projects

 Johané Nienkemper-Swanepoel (PhD)

 Visualization for categorical data with missing values

 Adriaan Rowen (Masters)

 Unravelling black box machine learning methods using biplots

 Raeesa Ganey (PhD)

 Principal surface biplots  Replace linear PCA with nonparametric manifold “following” the data

 Sasol Technologies R&D

 Ruan Rousseuw (PhD)  Visualisation for online process monitoring  André Mostert (PhD)  Machine learning in multivariate process control

Autor: 27.02.19

Applying machine learning to solve problems in remote sesnsing, trajectory mining and interferometry

T.L. Grobler tlgrobler@sun.ac.za

Autor: 27.02.19

Remote Sensing

Settlement detection The art of converting data about the Earth‘s surfice recorded with distant platforms into usable inforation.

Autor: 27.02.19

Trajectory mining

Submitted conference paper The automatic identification system (AIS) is an automatic tracking system that uses Transponders on ships and is used by vessel traffic services (VTS). Problems:

• 1. Trajectory mining.
• 2. Data representation.
• 3. Ship tracking.
• 4. Anonmoly detection.
• 5. Route Prediction.

Autor: 27.02.19

Interferometry

Connecting antennas together to form a single telescope whose purpose is to observe celestial radio emission.

Autor: 27.02.19

Typical Problems

Radio Galaxy Classification Burger Becker (17522021@sun.ac.za) Radio Frequency Interference Atenna failure analysis Lydia de Lange (18350070@sun.ac.za) Danie Ludick (dludick@sun.ac.za)

Autor: 27.02.19