AI and the Next Digital Frontier in Asia
Maithra Raghu - Google Brain
A deep representation analysis tool for learning dynamics, compression and interpretability
Deep reinforcement learning has achieved many recent successes, but our understanding of its strengths and limitations is hampered by the lack of rich environments in which we can fully characterize optimal behavior, and correspondingly diagnose individual actions against such a characterization. Here we consider a family of combinatorial games, arising from work of Erdos, Selfridge, and Spencer, and we propose their use as environments for evaluating and comparing different approaches to reinforcement learning. These games have a number of appealing features: they are challenging for current learning approaches, but they form (i) a low-dimensional, simply parametrized environment where (ii) there is a linear closed form solution for optimal behavior from any state, and (iii) the difficulty of the game can be tuned by changing environment parameters in an interpretable way. We use these Erdos-Selfridge-Spencer games not only to compare different algorithms, but also to compare approaches based on supervised and reinforcement learning, to analyze the power of multi-agent approaches in improving performance, and to evaluate generalization to environments outside the training set.
Maithra Raghu is a researcher at Google Brain and a PhD student at Cornell University. Her primary research interests are in better interpreting and understanding the representations learned by deep neural networks. Previous work has developed a technique (SVCCA) for comparing latent feature maps of convolutional networks and resulting in faster training methods. She has also worked on adapting a new testbed for deep reinforcement learning algorithms to enable studies of generalization, comparisons to supervised learning and multiagent performance.
Icarus So - TVB
Apply Reinforcement Learning to Programme Recommendation
Many people think reinforcement learning is too advanced and too academic which cannot apply in real business cases. The fact is just an opposite and very easy to be adopted gradually. In the presentation, Icarus will share his experience how to apply it in program recommendation from scratch to deep reinforcement learning. Traditional recommendation algorithms are collaborative filtering and content filtering. Icarus will explain how to encode them in a deep reinforcement network and add another important feature set, impression to the neural network. Difficulties such as avoiding local maximums and tricks on how to apply on startup will also be shared.
TVB is the largest television company in Hong Kong and its service spreads to Southeast Asia. It has more than million of users watching every day. Icarus is a Data Scientist in TVB. He designs the workflow and drives different data science projects which include program recommendation for MyTvSuper and BigBigChannel. Previously, he worked in Jobable, a startup matching candidates and job posts. He worked closely with the marketing team and help raise funding for series A with collaborated analysis blogs and deep learning models.
Johnson Poh - DBS Bank
Speech Enabled NLP for Event Extraction
It is often challenging to make sense of the information hidden within large volumes of unstructured data and speech. How do we seamlessly classify and extract key ideas with automation? In this presentation, we explore the open source tools, algorithms and services that may come in handy for the design of a reference architecture to sieve out underlying insights from texture descriptions. More specifically, we discuss how we can assemble a suite of text mining and probabilistic graphical processes into an automated pipeline, built on top of a modular framework to surface embedded topics of interest.
Johnson is currently Head Data Science for Big Data Analytics Center of Excellence at DBS Bank. He holds an adjunct faculty appointment at SMU School of Information Systems where his core focus areas include applied statistical computing, machine learning as well as big data tools and techniques. An avid programmer and data enthusiast, Johnson enjoys developing apps and data products. Most recently, he was awarded first prize in Singapore’s largest coding competition, Hackathon@SG 2015 as well as the CapitaLand Data Challenge 2016. Johnson completed his bachelor’s degree at University of California, Berkeley, majoring in the subjects of Pure Mathematics, Statistics and Economics. He received his postgraduate degree in Statistics at Yale University.
Nishank Singla - GE Digital
End to End Speaker Recognition System
Nishank is the co-founder and on advisory council of Alpha Brain Inc. Alphabrain speacialises in real world application of A.I. and help its clients to develop end-to-end solutions. Currently, Nishank is working as Software Engineer at Gneral Electric Digital. He received his master degree in Software Engineering from San Jose State University, San Jose California in 2017. He has been involved in various projects in A.I., deep learning and big data since 2015, named developed end-to-end face recognition system with autoface detection, clustering and recognition using deep learning models and OpenCV. He has developed speaker recognition system to make voice assistants as true personal voice assistants.
Pallab Maji - Mercedes-Benz R&D India
Optimizing convolutional neural networks for embedded platforms
It is widely known that neural networks applications are computational and memory intensive. Apart from the atomic computations of layers (convolutions, activations, fully connected layers, etc.) in the convolutional neural networks (CNNs), the base architecture of the CNN used in the inference is prominent in embedded platforms. Present ecosystem of the embedded platforms with respect to processing of CNNs are not completely matured yet, so as to be used as out-of-box solutions in a fully autonomous vehicles. Thereby, optimization of CNN is imperative for any embedded platform. This talk will first visit on various techniques in linear algebra that can help in achieving required reduction in computation and memory footprint. Then the various techniques that can actually reduce computational complexity of the CNNs. The talk will be focused on the optimization aspects of CNN for computer vision applications targeted for fully autonomous driving.
Pallab Maji is senior research engineer for autonomous driving at Mercedes-Benz Research and Development India at Bangalore. He leads a team of machine learning research engineers at MBRDI to work on interaction of autonomous vehicles with humans in the road through computer vision. He received masters’ and doctorate degree in electronics and communication engineering from National Institute of Technology, Rourkela India. His primary focus of research was on developing artificial intelligence algorithms like fuzzy logic and neural networks for various platforms like DSPs and FPGAs. His research interests includes machine learning, computer vision and embedded platforms.
Masashi Sugiyama - RIKEN Center for Advanced Intelligence Project
Machine Learning from Weak Supervision
Machine learning with big data is making a great success. However, there are various application domains that prohibit the use of massive labeled data. In this talk, I will introduce our recent advances in classification from weak supervision, including classification from two sets of unlabeled data, classification from positive and unlabeled data, a novel approach to semi-supervised classification, and classification from complementary labels.
Masashi Sugiyama received the PhD degree in Computer Science from Tokyo Institute of Technology, Japan in 2001. He has been Professor at the University of Tokyo since 2014 and he has concurrently served as Director of RIKEN Center for Advanced Intelligence Project since 2016. His research interests include theories and algorithms of machine learning (such as covariate shift adaptation, density ratio estimation, and reinforcement learning) and their applications to real-world problems.
Guan Wang - Swiss Re
A Very Brief End-To-End Overview of Chinese NLP
Guan is currently a Data Scientist at Swiss Re. His job involves using NLP technology for Automation and Augmented Intelligence. Previously Guan was working on Data Mining projects at Lenovo Machine Intelligence Center, and Machine Vision projects at ASTRI. During his Physics MPhil study at HKUST Guan was also doing research work on Genetic Algorithms and Complex Network Analysis. Guan trains himself to train machines.
From web crawling to annotation tools, from sentences to words characters and radicals, from embedding and weights to labels and losses, from LSTM and CNN to bi-direction and iterated dilation, from classification to sequential tagging and generation, from entity and relation to knowledge graph, from you to bots, here is a very brief end-to-end overview of Chinese NLP.
Intelligent Control and Cognitive Control over Robotics
Alleviating Skilled Labour Shortage with Automation
Joni Zhong - National Institute of Advanced Industrial Science and Technology, Japan
Cross-modal Understanding and Prediction for Cognitive Robots
Our brain works as a predictive machine, which works constantly as an active inference and keeps utilizing both active motor action and perception to minimize the prediction error. At the first part of the talk, I will briefly introduce this hypothesis in neuroscience and psychology and proposed a hierarchical predictive cognitive model for robots. At the second part, I will present how to apply this cognitive model to crossmodal understanding and prediction for robots using the state-of-the-art machine learning algorithms. A few robot video demonstrations based on these methods will also be shown.
Junpei “Joni” Zhong is currently a research scientist at National Institute of Advanced Industrial Science and Technology (AIST), Tokyo, Japan. He received BEng from the South China University of Technology in 2006, M.Phil from the Hong Kong Polytechnic University in 2010 and Ph.D. ("with great distinction") from the University of Hamburg in 2015. From 2014 to 2018, he has been participating in a few projects in Germany, UK, and Japan. His research interests are machine intelligence, machine learning, cognitive robotics and assistive robotics. He has been awarded the EU Marie-Curie Fellowship from 2010 to 2013. He is also a founding member of the organization "Consciousness Research Network" and a Guest Editor of journal "Complexity" and "Interaction Studies".
The Rise of Artificial Intelligence and the Threat to Human Rights
PANEL SESSION: Who Should Be Responsible for Regulating AI?
Juergen Rahmel - HSBC Germany
Dr. Juergen Rahmel received his PhD (Artificial Intelligence) in Germany for research which covered Artificial Intelligence and Machine Learning in the areas of Case-based diagnosis, connectionist learning with Neural Networks as well as topological and hierarchical interpretation of Neural Networks. The results of his research were successfully applied in the fields of medical diagnosis (nerve lesions, tumor grading) as well as technical diagnosis (fault detection).
Dr. Rahmel has extensive experience in the international financial sector, in roles as Director, Head of IT and Programme Manager of global initiatives for multinational banks. He served as the Head of Artificial Intelligence in the Applied Innovation Team of HSBC as well as Head of Innovation Quality management for all areas of FinTech technologies. He currently is Chief Digital Officer of HSBC Germany.
Kristi L.Swartz - Bryan Cave
As Bryan Cave’s Hong Kong office managing partner, Kristi Swartz concentrates her practice in corporate finance, focusing on SFC investigations and fintech matters such as robo-advisory services, peer-to-peer lending and alternative payments. Kristi has practiced in Hong Kong for over 20 years, and works with emerging growth companies through to listed multinational corporations in Hong Kong, China, Asia and Europe across a broad range of industries. Recognised as a leader in the financial technology and regulatory fields, Kristi has spoken on panels providing insight for companies to remain compliant and was named one of the “Women in Finance” by Let’s Talk Payments. Kristi is a front-runner in bringing new concepts to the market, and is currently advising on the regulatory requirements for ICOs and the implementation of equity crowdfunding for professional investors.
CONVERSATION AND DRINKS
REGISTRATION & LIGHT BREAKFAST
Tak Lo - Zeroth.AI
Tak Lo is creating startup leaders in Asia. He is a Partner at Zeroth, an AI-focused early stage funding program. Previously, he sourced, invested, and mentored companies in the $3m Techstars London seed fund. He has worked with over 50 companies from the Asian, European, and NYC startup communities angel investing, founding, and mentoring startups. Tak started his career in the US Army. He graduated from the London Busi ness School and the University of Chicago.
Does Machine Intelligence Need Emotional Intelligence
Machine Intelligence for the Next Generation of Education
Personalise Your Trip with Machine Learning
Creating Art or Reproducing Art
Kendra Vant - Seek Limited
Why Settle? Using Machine Learning to Help People Find New Opportunities
When you search for a job, you want to know you have seen everything available that's right for you. If you're looking for your next team member, you want to find the standout applicants as quickly as possible. With 150 million candidates and 700k hirers across 16 countries, the scale of Seek's search and match problem long since outstripped any possibility of manual curation of opportunities and candidates.
In this talk I will introduce our ongoing work using machine learning to build radically more efficient and effective employment marketplaces. In particular, I will discuss our recent efforts combining natural language processing and Deep Learning to connect people to their dream job.
A firm conviction not to die wondering ‘what if?’ has given Kendra Vant a rich and varied career working with memorable people, companies and universities across New Zealand, Australia, the US and Malaysia. Through it all, her greatest satisfaction has come from working with smart people to solve difficult problems.
After doctoral research in experimental quantum physics at MIT and postdoctoral work in applied quantum computing at Los Alamos National Laboratory, Kendra was serendipitously placed to ride the tsunami of corporate interest in applying machine learning to create personalised experiences in an increasingly connected and digital world.
She has worked in insurance, banking, telecommunications, government, gaming and the airline industry and is currently Principal Data Scientist with Seek Asia Pacific and Americas, applying emerging techniques in natural language processing and deep learning to the problem of finding other people their dream job.
Hamed Valizadegan - NASA Ames Research Center
Machine learning For Space Projects: Example Engineering and Science Case Studies
The success of the space projects depends much on our ability to understand and analyze their collected science and engineering data. In the science domain, the amount of collected data is so very large that requires building automatic tools to make sense of them. In this domain, often the data is not annotated well and/or there is not enough representative features for effective model construction. In the engineering domain, there are components that are designed and built for the first time and there is a limited domain knowledge available for them. This makes the hand-coded physics-based models less accessible and models that can utilize data for prediction more desirable. However, often there is a small number of working and failed units to learn from.
In this talk, I provide examples of both cases and show how machine learning can help in these imperfect scenarios. First, I use our efforts in life time prediction of Fine Guidance Sensors of Hubble Space Telescope as an example to demonstrate how little domain knowledge can help us develop effective machine learning models when the data availability is scarce. And then, I demonstrate our experience of using machine learning to classify the transit-like signals of Kepler spacecraft when annotation is imperfect and data features are not representative.
Holder of a PhD in computer science with focus on machine learning and data mining from Michigan State University, Hamed Valizadegan joined NASA Ames Research Center (UARC) as a machine learning research scientist in 2013. At Ames, he has been involved with multiple projects including Hubble Space Telescope, Kepler mission and ASRS aviation safety reports! Before joining NASA Ames, he spent three years at University of Pittsburgh conducting research in Medical Informatics. He has published in prestigious venues such as International Conference on Neural Information Processing Systems (NIPS), ACM SIGKDD conference on Knowledge discovery and Data Mining (KDD), and Artificial Intelligence and Statistics (AISTATS).
Vidal Fernández - China Light & Power Holdings
Using Big Data & AI to Optimize Wind Farm Performance
Big Data & Streaming Analytics Platform help us to collect real time data from IoT sensors that capture wind data and store in our Data Lake. Conventionally, every wind turbine in a wind farm is operated to maximize its own power production without taking into account the interactions between the wind turbines in a wind farm. Because the wake interference, such greedy control strategy can significantly lower the power productions of the downstream and reduce the overall wind farm power production. Using AI methods, based on data collected, we built and deployed a set of AI algorithms to optimize the overall wind farm performance. Based on every wind turbine optimization and the interactions we optimized the overall wind farm performance.
Dr. Fernandez as Director of Big Data Analytics at CLP Holdings is developing the big data strategy with focus on innovation and emerging technologies. Dr. Fernandez is leading the intellectual process of researching and analyzing new technologies, processes and businesses, gathering and synthesizing information and formulating insightful findings and conclusions to deliver value. He is managing a talented team of data scientists to ensure we are deriving from data the right insights to improve business processes, being more efficient, identifying new business models and reshaping the existing ones. Dr. Fernandez has developed several BIG DATA & Analytics solutions as AaaS, based on cloud, for many blue chip companies in different industries: Banking, Insurance and financial Services, Telecommunications, Retail, Utilities and Oil & Gas,... Customer Insights, Smart Grid Analytics, Customer Experience, Predictive Asset Maintenance, Operational Efficiency and Risk Management, Fraud Detection, Anti Money Laundering, Price Optimization, and many other business challenges are some of the delivered solutions deployed.
Mohammad Yousuf Hussain - HSBC
Application of Generative Adversarial Networks (GANs) in Algorithmic Trading and Aggregation of Low-Alpha Strategies
Machine learning is attracting a lot of attention recently due to the promising capabilities of adapting, learning and even self-teaching algorithms. Application of machine learning in algorithmic trading offers a variety of opportunities to extract greater alpha and improve order execution. In this session, I am going to present the application of Generative Adversarial Networks (GANs) in algorithmic trading and share insights on aggregation of strategies.I would highlight the key functions of a strategy aggregator; categorisation of strategies, classification of market behaviour and application of ensemble learning. I would then briefly discuss about the design and development of controls for the aggregator framework that allows ease in monitoring and optimisation of the solution.
Mohammad Yousuf Hussain, CFA is a Senior Technology and Innovation Specialist at HSBC. Working in the Applied Innovation and Strategic Investments team, he has designed and delivered a number of artificial intelligence based electronic trading solutions. Previously, he was a Senior Consultant at GreySpark Partners where he delivered projects for UBS, HSBC, Mizuho Securities, Nomura-Instinet, Interactive Brokers and SFC. He developed expertise in assessing trading algorithms by investigating the market abuse incidents for the regulator.
Martin Qiao - HSBC
Banking operation automation with AI techniques
Many financial service companies are at their early stage to embrace Artificial Intelligence into their daily operation to improve efficiency and reduce human intervention. During this journey the most common struggle is to swing between build a home-brew solution versus buy a vendor solution. In my presentation I would like to share my experience in introducing Artificial Intelligence applications into banking operations, and trying to share my views on when to choose vendor products and how to assess them, and when to develop internal solutions and how to chain up open sources packages into a industrialized solution. Chatbot and sanction screening will be introduced as use cases in terms of NLP area, and signature verification solutions in terms of computer vision area.
Dr. Martin Qiao is currently working as the Lead Architect for Machine Learning Automation in HSBC. He is holding a PhD degree for Mathematics with research experience on various image-related processing. His working experience is mixed with Data Science for insurance marketing and luxury marketing, NLP and computer vision solution design for banking.
Rajan Kumar Upadhyay - DHL
Customer Engagement with Omni/Multi-Channel AI Chatbot
Rajan Kumar Upadhyay is the Machine Learning Technical Expert at DHL, He holds PGPX from UCLA Anderson(US) & APSM from IIM Calcutta(India). He is specialist technologist and togaf 9 certified enterprise architect. he has more than 16 years of strong IT experience varies from various platforms to various technologies. He owns a consistent track record of increasing market share, organizational potential, profitability, and product value within the global markets. In his current role he is associated with DHL, helping supply chain business transformation using intelligent combinations of Data Science , Machine Learning, Natural language processing & Robotic process automation. His day always starts with thinking something new & rolling out those ideas into concept & then reality.
Danfeng Li - Alibaba
User Behaviour Data Is a Gold Mine, and How We Dig It
User's online and on mobile data provide an unique perspective to understand customer's behavior. When the behavior is directly related to your business, it's easy to understand the impact if the data is correctly applied. However, in some cases, seemingly non-related behavior can also be found very useful through machine learning and data mining. In this talk I'll give two specific examples, namely internet fiance risk control and online advertising, to show how sophisticated models can find the underlying correlation between business goal and online and on mobile behavior.
Dr. Danfeng Li is currently a director in Alibaba group, and the Chief Data Officer of Umeng+, a company fully owned by Alibaba. Umeng+ provides analytic tools for web and app developers. It helps more than 1.45 million apps and 7 million websites monitor and management their data. Umeng+ also offers serives based on its online and on mobile data. Dr. Li leads the effort to apply data mining and machine learning technologies to help clients in retail, finance and other industries to understand and management their customers. Before he joined Alibaba, Dr. Li was a principal development lead in Microsoft. He also worked at Opera Solutions, Yahoo and Fico. Dr. Li received his Ph.D. degree from University of Illinois at Urbana Champaign, and bachelor degree from Tsinghua University.
Yu-Xi Chau - A.S. Watson Group
A Flexible answer to Retail CRM: an Ensemble engine for Customer Management
An ensemble model is built on a Spark architecture that utilizes a range of models, including Gradient Boosted Trees, multi-layer Neural Network, Collaborative Filters and others. We show that by grouping the outputs of individual models and collecting their ensemble results from weaker models, rollout to multiple customer management use cases is extremely efficient and saves costs by 70%, yet preserving model robustness and reduces overfitting.The forever complex and versatile nature of retail requires a highly adaptive model with a robust ability to incorporate complicated and changing business rules. Therefore this paradigm allows us to collect and append past heuristics easily, and allow rapid rollout to multiple channels and business units.
From a very young age, Yu-Xi has always been motivated by the thrill of solving problems. This has led him to earn a doctorate in Complexity Science and subsequently entering the industry to solve business problems with Artificial Intelligence. Since then, He has tackled a range of problems, ranging from Gaming, Election Engineering and Retail. Apart from delivering products to bring business value, he has also built up Data Science Teams, training, recruitment and setting best practices. He is currently heading the Data Science Team in A.S. Watson, the biggest Health and Beauty Retailer in the world.
END OF SUMMIT