“There are a lot more papers written than there are widely read!” (Tom Mitchell)
Prof. Tom Mitchell is one of the giants of machine learning and artificial intelligence, who started the world’s first machine learning department at Carnegie Mellon University (CMU). Prof. Mitchell has kindly agreed to an interview with us at MLDawn, presented below with a list of the topics discussed and their corresponding timestamps in the video:
00:30 A brief introduction about Prof. Tom Mitchell in his own words
01:08 How did Prof. Mitchell become interested in the field of machine learning?
03:00 The current research interests of Prof. Mitchell: Conversational Learning
06:57 The famous Machine Learning book of Prof. Mitchell
10:24 The 2 continuous learning agents named NELL and NEIL developed by Prof. Mitchell and his team: How long have they been learning, and what have they been learning?
19:49 The gap between Real Neural Networks and Artificial Neural Networks and how to make the gap disappear?
26:35 The fairness of current reviewing process in conference venues belonging to big names in machine learning
30:25 Just using readily available Machine Learning libraries (e.g., Pytorch, Tensorflow, etc.) vs. understanding the details under the hood as well!
A Summary of the Main Points in the Interview
Prof. Mitchell is one of the giants of machine learning and artificial intelligence who started the world’s first department for machine learning at Carnegie Mellon University (CMU). He has always been interested in the concept of “learning” and has had a lot of different majors, including psychology. However, in the end he felt it did not provide enough observability of what goes on in humans’ brains, so he changed his direction towards studying intelligence in computers. However, after the advent of brain imaging, he made connection with a professor in the department of psychology at CMU who was working on brain imaging at the time. It was then that Prof. Mitchell focused on intelligence in computers and brain activities during intelligent processes in humans.
Prof. Mitchell is currently highly interested in conversational machine learning: where machines will learn from instructions, demonstrations, and basically conversations. At CMU, Prof. Mitchell and his team have developed 2 continuous learning agents called Never Ending Language Learner (NELL) and Never Ending Image Learner (NEIL). They both look at the web and try to extract more structured information. For example NELL has over 100,000,000 beliefs, for instance, Roses represent the emotion of love, Bill Gates founded Microsoft, etc. Prof. Mitchell explains how lack of the ability of self-reflection by NELL, made him pull the plug and stop NELL’s training after nearly 8 years! Prof. Mitchell and his team have explained the idea behind NELL and the lessons learned from that experiment in a publication at CACM in here. Prof. Mitchell believes that there are main differences between an actual biological neural network and an Artificial Neural Network (ANN). His team and he, have published a paper to address the connection between biological and actual neural networks: Predicting Human Brain Activity Associated with the Meanings of Nouns. Prof. Mitchell believes that the reviewing process at machine learning conferences/journals is biased due to humans’ imperfection and also a systematic bias for well-known people and institutions.