The art and science of applying machine learning techniques inside for a profit company is a world away from pursuing algorithm improvement and fundamental in a research setting. I will talk about the end to end process of building smart products within a SaaS company today.
Kendra is an industry leader in building data-driven products by harnessing emerging artificial intelligence and machine learning techniques to solve problems for businesses and industry globally.
She has had a rich and varied career working in New Zealand, Australia, the US and Malaysia, leading data and engineering teams at companies including SEEK, Telstra, Deloitte and now Xero. At Xero, she heads a globally distributed team of developers, machine learning specialists and data practitioners using emerging practices and technologies to make data work harder for small businesses and their advisors.
After doctoral research in experimental quantum physics at MIT and postdoctoral work in applied quantum computing at Los Alamos National Laboratory, Kendra worked in bespoke software development and then in generating business insights from data before focusing on applying machine learning to create personalised experiences in an increasingly connected and digital world. She gets her greatest satisfaction from working with smart people to solve difficult problems that have a positive impact on the world.
Most NLP approaches use external language resources, such as text corpora, to derive the distributional properties of word usage and represent linguistic meaning. In this talk I will review work from cognitive science exploring to what extent linguistic meaning depends on other factors as well, and how to capture them computationally. In the first part of the talk I will compare standard word-embedding models derived from corpus data to a semantic network derived from an extensive dataset of word associations involving more than 12,000 cue words and over 500K participants. I'll demonstrate that the word embedding model fails to capture important aspects of people's lexical representations that are captured by the word-association-based semantic network -- aspects which probably reflect environmentally-grounded sensory knowledge as well as pragmatic and emotional understanding. In the second half, I will review evidence suggesting that human language learning involves active exploration and sophisticated conceptual/social reasoning in addition to bottom-up distributional mechanisms. Implications for NLP and computational linguistics will be discussed.
Andrew Perfors is an Associate Professor in the School of Psychological Sciences and Deputy Director of the Complex Human Data Hub at the University of Melbourne. Andy's research involves combining computational models and controlled experiments to better understand higher-order cognition. His research topic concerns how people's cognition interacts with the social and linguistic environment, with each being shaped by and shaping each other. In particular, he explores how linguistic and conceptual knowledge is mediated by (and mediates) complex statistical and social reasoning grounded in a rich and complicated environment. You can check out more at his website or by following him on twitter.