Color me intrigued -- it certainly does seem interesting, and potentially unifies symbol and logic based GOFAI with the "throw more statistics at the problem" school that currently encompasses "AI" as a buzzword. If we're still on for August 15 I would love to attend as I think I would learn much.
Bikeshed issue: You may wish to consider revising the title of your talk. "Make X great again" is considered "normalizing hate speech" and very offensive, therefore unacceptable by 2019 professional standards. Major open-source conferences will reject a talk proposal with such a title, and some may preëmptively ban you from the con entirely for proposing it. Source: https://mobile.twitter.com/aurynn/status/1128053124323655680
It may not be an issue among us (Faré may even like it, and relish a twist of the old knife), but among less laid-back groups and conferences it can be severely career-limiting.
On Mon, Jul 29, 2019, 7:33 AM Masataro Asai guicho2.71828@gmail.com wrote:
Thank you for the invitation :)
August 15th should work, but I need to confirm the schedule in the company, which I cannot check now until I complete the onboarding process of US IBM (although I have been in IBM Japan for quite a while...)
Will they be posted on any sort of MIT bulletin board? Is it more of an academic event? That might need a permission from the manager. For an informal meeting it should be fine.
Blurb meaning the short description / abstract of the talk? Let me give a try...
For myself:
Masataro Asai, aka guicho271828, is a Lisp hacker with 11 years of experience in Common Lisp, who received a Ph.D in Artificial Intelligence from the University of Tokyo. He is an expert in heuristic graph search and automated planning and scheduling, also known as AI Planning, with publications records on top AI conferences e.g. AAAI, IJCAI, ICAPS. Author of a fast pattern match library Trivia, a numerical computation library NUMCL (which got 400 stars) and more.
For the talk:
Title: Make Symbols Great Again!: Classical PDDL/STRIPS Planning in Deep Latent Space
Domain-independent classical planners require the symbolic models of the problem domain and instance as input, resulting in a knowledge acquisition bottleneck. Meanwhile, although deep learning has achieved significant success in many fields, the knowledge is encoded in a subsymbolic representation which is incompatible with symbolic systems such as planners. We introduce LatPlan, an unsupervised architecture combining deep learning and classical planning. Given only an unlabeled set of image pairs showing a subset of transitions allowed in the environment (training inputs), and a pair of images representing the initial and the goal states (planning inputs), LatPlan finds a plan to the goal state in a symbolic latent space and returns a visualized plan execution.
In the talk, we also survey the recent progress in the related fields and also informally talk about my future ambitions.
Regards Masataro
Faré wrote:
Congratulations!
Let's summon a Boston Lisp Meeting to greet you. How about Thursday August 15th at 1800?
I'll try to reserve a room at MIT. Either way, we can meet for dinner afterwards.
Can you give a speech on some topic of your choice? Otherwise is there any other candidate? Can you give me a one-paragraph blurb about yourself, another one about your topic?
—♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics•
Genius is one percent inspiration and ninety-nine percent perspiration. — Thomas Alva Edison
On Sun, Jul 28, 2019 at 7:27 AM Masataro Asai guicho2.71828@gmail.com
wrote:
Dear all,
I just moved to Boston yesterday. 5yr assignment. Thanks
-- Masataro Asai Ph.D
Research Staff Member IBM Research
Tel: +81-44-856-9009 Mobile: +81-50-5534-1357 Mail: guicho2.71828@gmail.com Website: http://guicho271828.github.io/
-- Masataro Asai Ph.D
Research Staff Member IBM Research
Tel: +81-44-856-9009 Mobile: +81-50-5534-1357 Mail: guicho2.71828@gmail.com Website: http://guicho271828.github.io/