When: Wed, 3 Apr 2019, 2pm
Where: A226, College Building
Who: Derek Doran,Wright State University.
Title: Mappers and Manifolds Matter!
Abstract: Topological Data Analysis (TDA) is a branch of data science that estimates and then exploits the “shape” of a dataset for downstream characterization and inference. TDA methods arerising in popularity in the ML community as a tool to theoretically understand the actions of deep neural nets and other algorithms by connections to the Manifold Hypothesis. TDA methods, and in particular the Mapperalgorithm, is also finding recent increased use in applied data science workflows. This talk will introduce the essential definitions and notions from Topology needed for audience members to jump into the field of TDA and to build Mappers of datasets. It will then demonstrate the utility of Mapper, and TDA methods generally, for data science tasks including complex data visualization, interpretable dimensionality reduction, and explainable deep learning. The application of Mapper for dataset characterization and semi-supervised learning will also be illustrated.
Bio: Derek Doran is an Associate Professor of Computer Science and Engineering at Wright State University, Dayton OH, USA. His research interests are in machine learning and complex systems analysis, with current emphasis on topological methods for explainable AI, deep learning, complex network analysis, and web and geospatial systems mining. He is an author of over 75 publications on these topics, four of which have been recognized with best paper awards and nominations, is an author on multiple patents, and has published a book under the Springer Briefs in Complexity series. He serves on the program committee of major AI and Web conferences, is on the Editorial Board of Social Network Analysis and Mining and has served as process improvement chair at ESWC. Dr. Doran is a National Science Foundation EAPSI Fellow, a graduate research awardee of the U.S. Transportation Research Board, and a twice summer alumnus of Bell Labs. He will be a Fulbright Fellow stationed in Reykjavik University effective January 2020. Please see more information at https://derk–.github.io.
Abstract: We introduce DeepProbLog, a probabilistic logic programming language that incorporates deep learning by means of neural predicates. We show how existing inference and learning techniques can be adapted for the new language. Our experiments demonstrate that DeepProbLog supports both symbolic and subsymbolic representations and inference, 1) program induction, 2) probabilistic (logic) programming, and 3) (deep) learning from examples. To the best of our knowledge, this work is the first to propose a framework where general-purpose neural networks and expressive probabilistic-logical modeling and reasoning are integrated in a way that exploits the full expressiveness and strengths of both worlds and can be trained end-to-end based on examples.
Bio: Robin Manhaeve is a PhD student at the department of Computer Science at the KU Leuven. In 2017, he completed his MSc in Engineering Science: Computer Science at the KU Leuven. He’s currently researching the integration of Deep Learning and Probabilistic Logic Programming under the supervision of Prof. Luc De Raedt and is funded by an SB grant from the Research Foundation – Flanders (FWO).
Who: Charitos Charitou; City, University of London
Title: Deep Learning for Compliance: “Application of machine learning to online gambling data to identify money laundering”
Abstract: Most of the current online gambling operators are using handcrafted basic rules for their anti money laundering (AML) strategy. These methods are not enough anymore for identifying complex fraudulent activities. Kindred group entered into research collaboration with City University and the main goal is to effectively use machine learning to detect money laundering. Understanding the needs of the industry and what the industry stakeholders believe was a priority. A series of interviews with various stakeholders of the gambling industry took place and the findings were published earlier this year in the form of a white paper.
The second part of the research involved the analysis and evaluation of the gambling data that were provided by Kindred. We present how the imbalanced dataset problem was tackled, and the new experimental dataset that was created for supervised learning. The performance of Logistic Regression (LR), Random Forest (RF) and Multilayer perceptron (MLP) was examined and compared. Our results, showed that Random Forest was the best model for predicting the normal players, while the MLP managed to detect suspicious players with the higher accuracy. Finally, the sequential relationship of the data was investigated using discrete and continuous Hidden Markov Models (HMM).
When: Wed, 20 Feb 2019, 2pm
Where: AG21, College Building
Who: Dr. Alberto Ferreira De Souza, Universidade Federal do Espírito Santo (UFES), Brazil.
Title: Building IARA – The Intelligent Autonomous Robotic Automobile
The Intelligent Autonomous Robotic Automobile (IARA) is one of the most advanced self-driving cars in world, figuring in eighth place according to the metric of number of interventions per 1000 miles in 2017. We start building IARA in 2009 and, since then, more than 30 students, including Ph.D., M.Sc., and undergraduates, have completed their courses working on the project. IARA is based on the precise localization paradigm where the self-driving car must have a detailed map of the environment to operate autonomously. In this talk, we will present some of the history behind the 10 years of research that led to the current status of development of IARA and will describe how some of its main software modules work, including the modules responsible for mapping, localization and autonomous navigation. A demonstration video can be found here.
Dr. Alberto Ferreira De Souza is a Professor of Computer Science and Coordinator of the Laboratório de Computação de Alto Desempenho (LCAD – High Performance Computing Laboratory) at the Universidade Federal do Espírito Santo (UFES), Brazil. He received B. Eng. (Cum Laude) in electronics engineering and M. Sc. in systems engineering and computer science from Universidade Federal do Rio de Janeiro (COPPE/UFRJ), Brazil, in 1988 and 1993, respectively; and Doctor of Philosophy (Ph.D.) in computer science from the University College London, United Kingdom in 1999. He has authored/co-authored one USA patent and over 130 publications. He has edited proceedings of four conferences (two IEEE sponsored conferences), and is a Standing Member of the Steering Committee of the International Conference in Computer Architecture and High Performance Computing (SBAC-PAD).
Abstract: With the inclusion of renewable energy into power systems, traditional power system face new challenges. Due to their inherent fluctuations and variability, the introduction of renewable energies in power systems poses new challenges in modeling uncertainty. Controlling and optimizing the operation cost by adjusting the output generation of renewable energy resources make power systems operation more reliable and secure. In this work, we aim to solve one optimal microgrid management problem in deterministic and probabilistic framework. This microgrid is connected to the utility and comprises of different renewable energy generators such as photovoltaic (PV), wind generators, batteries, hydroelectric plants, and microturbine.
The objective is to minimize the cost of generation and the voltage deviation from the reference.
The optimization problem is nonlinear since the AC load flow which is a constraint in the optimization problem is nonlinear. we linearize AC load flow in the first step. Secondly, we model the problem in a deterministic framework without considering the impact of uncertainties on power system. The other physical constraints which are taken into account in this work are the equality constraint of load generation balance, output power limitation and voltage limitation. Finally, we will model the nonlinear problem in probabilistic framework to see how uncertainties can affect the system. The first two steps have been done before transferring to PhD and the final step will be done in the next following two years.
Tillman Weyde, Rahda Kopparti, Dan Philps and Artur Garcez attended and presented papers at NeurIPS 2018 in Montreal, Canada, during the week of 3 Dec 2018. Dan and Artur provided an informal overview of their impressions of NeurIPS to the ML group’s End-of-Year meeting on 14 Dec 2018. Thanks to Benedikt Wagner for organising the meeting.
Luciano Serafini (FBK, Trento, Italy) and Michael Spranger (Sony CSL, Tokyo, Japan) visited City’s Research Centre for Machine Learning during the week of 10 Dec 2018. The main focus of the visit was to continue research collaborations on Logic Tensor Networks. LTNs are a deep learning system implemented in Tensorflow, capable of reasoning with first-order many-valued logic. For more information, please check the webpage of the IJCAI’2018 tutorial on LTNs: https://sites.google.com/fbk.eu/ltn/tutorial-ijcai-2018.
Artur Garcez gave a lecture on Relational Neuro-Symbolic AI at the EurAI Advanced Course on AI, 2018, which took place in beautiful Ferrara, Italy.
All the lectures, with overarching theme Statistical Relational AI, are available from the University of Ferrara’s YouTube channel: https://youtu.be/KeFhKi-tOTs?list=PLJPXEH0boeNDWTNwWTWnVffXi5XwAj1mb
Artur Garcez gave two talks: Part 1 gives an overview of two decades of research on neuro-symbolic AI. Part 2 describes in some detail two neuro-symbolic systems for relational learning: Connectionist ILP and the Logic Tensor Networks framework.
Prof Paul Smolensky, Johns Hopkins University and Microsoft Research, will be visiting the Research Centre for Machine Learning for a conversation on neuro-symbolic computing, his book (with Legendre), The Harmonic Mind, MIT Press, and recent papers and results using Tensor Product Representations (TPR) and learning such as A neural-symbolic approach to design of captcha, https://arxiv.org/abs/1710.11475, as well as applications of TPRs to natural language tasks and data sets such as the Stanford question answering dataset SQuAD.
City, University of London is an independent member institution of the University of London. Established by
Royal Charter in 1836, the University of London consists of 18 independent member institutions with outstanding
global reputations and several prestigious central academic bodies and activities.