sustaining humans? the search for transformation of university ..... connections between soros ,fazle abed ,crow and botstein-all are founders of the coalition OSUN -open society university networking - soros founded OSUN - open so ...............................................................................................................................................................................................................
Tuesday, September 1, 2020
twiml suzanna llic ml tokyo and caualy london
twiml uni updates 20 oct 2020
Milind Tambe
Director of AI for Social Good
Google Research India
Milind Tambe is the Director of AI for Social Good at Google Research India. His research focuses on advancing AI and multiagent systems research for Social Good. Milind contributed several foundational papers in Artificial Intelligence in areas of agents and multiagent systems and computational game theory. This research has led to significant practical impact, such as use of the green security games framework to assist wildlife conservation around the world, use of social networks and machine learning to assist in improving public health outcomes such as HIV prevention, and the use of pioneering security games research for security optimization by agencies such as the US Coast Guard and the Federal Air Marshals Service. Milind is also a Professor of Computer Science and Director of the Center for Research in Computation and Society (CRCS) at Harvard University.
In this session, Sam is joined by Milind Tambe, Director of AI for Social Good at Google Research India, and
35 years on agency ai
more recently focus ai impact
grew up mumbai - attacks india mother in trin attacked
these events sharpened interestpublic safety airoprts etc 15 years ago
randomisecheckpoints with gaetheory eg lax
now in wildife and public heallth
relation appn of ai and context of social good? in my view social impact work can advance ai into new areas that dont come up wotherwise
from 2009 we have been struggling with big data when no data so data has to be collected/standard ai techniques only start when perfect data exists
community data socially has to be interdisciplinary
..eg conversation scientists, social workers on ground, linking together separated ngos, understand ngos challege on ground- improve use of those resources that they have- we are not married to the tool but start with the social problem
eg youth homelessness - problem brought to us by ngos
-6000 yout sleep strrets lo angeles 13-24 - rates oh hiv i yhis population 10 tomes norm - so how speadinfo to these 6000 youth- map who are the key networkers - brief them- can we improve on this model
key nodes in network of the 6000 -somewhat like viral marketung bt now with life critical knowhow- ai algoritms finddiffeent youth than ngo asking wh is most popular?
- man finding - diffeent youth because alogorith lookedfr strategic communities - eg beach bum network, basketball network - straegicallyplaced youth different fro mass
-twist - compute sciece sume social netwrk id given- this makes sense when marketig so not having the network data to start with - twist has to do earlier research
challenges
Sam Charrington
@Anupam Ghosh your guides were overfitted
Milind TambeYes, in Google Research India, one project is focused on how to get dams to operate better to avoid floods, and ensure we provide enough water to farmers etc
The TWIML AI Podcast with Sam CharringtonGreat question Dinesh.
Milind TambeImproving dam and barrage water release: Using AI to inform dam and barrage water
Anupam Ghoshyeah most likely - real nueral nets
Milind TambeThis is obviously just one project, but the challenge you are pointing to Dinesh is a very important and massive one!
Sam CharringtonMilind, would some of your work on poaching be applicable to illegal fishing? I recently saw a story about an increase of poaching of the coast of the Galapagos islands.
The TWIML AI Podcast with Sam CharringtonHow would it translate?
Milind TambeAbsolutely! I agree that would be a great application for fighting illegal fishing
Anupam Ghoshare these learnings (ai models) from a particular region transferable to other regions? e.g. healthcare projects which you are working on.
Sam Charrington
We see a lot of this already, e.g. recommendations, algorithmic feeds, clustering.
Dinesh ChauhanIn general how facts can be separated out from fake news.
Milind TambeYes, there are some key problems that across areas. For example, how to spread health information...whether its HIV, or TB. Whether its the US or India, the problems are similar.
Sam Charrington
Mmm. tough problem. Have you worked in this area @Milind Tambe ?
Milind TambeSo there is some commonality across domains in problems, but providing tools that would be usable by non-profits directly would be absolutely awesome...thats a great problem to be working on
Anupam Ghoshit's surprising to see similarities between US and India in this context.
chris Macraedoes ming work with iqbal quadir at harvard - he started collecting village phone data 25 years ago- lots of deep problems and ngos
Anupam GhoshI guess people/behaviour has some universality
Milind TambeWe have done preliminary work in how in a social network we could try to neutralize misinformation but there is just a lot that needs to be done, e.g., as we come to COVID-19 vaccine.
Anupam Ghoshlaw
Milind TambeWe will need to combat misinformation about the covid-19 vaccine
Milind TambeSpots for illegal fishing would be similar to spots for poaching
india health., conversation paired up india ngos and tech colleaguesmaternal child care preventing hiv bangalore, tb, pandemics see blog
Thank you to our confirmed presenters:
During discussion you should be prepared to share code if relevant but not run it. The goal is to provoke interesting project related discussion to help the presenting and audience learn and progress their knowledge and skills. Any and all ML topic areas are welcome.
MLT. Machine Learning Tokyo. General information: Founders: Suzana Ilic, Yoovraj Shinde Founded in: July 2017, Tokyo, Japan Registered nonprofit 一般社団 ...
MLT(Machine Learning Tokyo) is a Tokyo-based nonprofit organization dedicated to democratizing Machine Learning. We are a ... Suzana Ilic. Founder and ...
moved back to london from master nat langage -applied lingisusistics - processing in tokyo-deep learning national institure= the contract for google japan
started non profit mlt to help 7000 members get into deep learning
now in london -causally conpany health ai
biomedica causal relationships
causal strength - genes chemics diseases
directionality of causal relationship eg a increases b didffebt a asociated with b
engine search huge data on possble ause/effects in just a few clicks
what might take weeks to research traditionally now tales secnbds to click
also vusual tool for how we interpt knowledge system intuitively
-how this expands precepts of nlp
one use case pre drug - what are risks of deigning drug in this area
we have domain experts in different drug subsegemnts
unique computtional linguist space - can be gamechanger in pharma profession
course with robert ness on causal model and machine learning
-- how did 3 of you hack mlt coomunity
we were 2 domain experts sortware education engne- me machine lingists
stared at co=work space - more and more joined us later workshops to cotrain each other
one half years ago turned from hack space to a non profit wanting to attract educators - 250 workshops panel discussions
groeing oyt of japan to india to canada europe to usa
evoilved from slf-learning passion to a global alumni netwrk
gamechanger met entreprenur tuzosan - founer of elita for kids
ai projects for scial good -nbdeep learning hardware- resarch publications nlp
egai efga -edge devices
edai lab umbrella
-became our contribution for socialgood
began collaboration chinba hacker community farmes and tech
sgri robots
this led to ajilab competition -traffic data tokyo object tracking system -retina nets
in mlt love collaborating with unis open science is an idea tokkyo institute waed to join in - 2 day bootcamp - intro machine learning to expert faculties
nlp machinelearningcreativity
diffeent pretraining methods
targeting oetry lyrics creativ space
edgeai project - machine learning for developing world - edge devices can adapt to frugal envoironments
in dev countries food productio in resource constrained places viutal- dep learning and edge devices
community-driven independent research is very important - a passion of mlt
interview andreas massen - ind researcher- main advice - try harder to find affilition
question can community provide support that an institution can provide?
in between open self-org and closed institution
hugging face democratising nlp
companies do open source nspire me
davy cadf at google brain - unique approaches machine learning
lear collab that people want to multiply and co-code
many approaches to causal modeling
explicit and implicit relatinships/architectures
more on transitoio dmain xpert to machine learning
mu unusial background - needed to learn both hands on and theorists but i am stll a lingist not compt science understanding the property of labguafe as a data
i define prblems as a lingist'then i look for prototype alogorithms
lookat what model len
then reiterate
steategic langage decisions
lingust as dmin expert
emily bender interview are ther ebnough linguists in nlp
lingisutsand machine learning gohand in hand
understanding impact of the model
at begiining of howto collab brig disciplines together
product owner work starts before tech analysis- you set targets for products impacs and communicate to team
maximise value for our users
ui user emperience
design principles of ai when statistical core
ui and ux
there are things which ae language specidc about causality but in london i am working in english
language though reaoning brin - lingusitics helped me understand that computer science wouldnt hsve - appled lingistocs - argumentation theory etc
solving language problems
mahine learing research and in production - 2 opposites see publication
The TWIML AI Podcast with Sam CharringtonDon't forget to get your questions for Suzana in!
Alisher Abdulkhaev[message retracted]
Alisher AbdulkhaevQuestion to Suzana: "Who had inspired (or still inspiring) you a lot from AI/ML community?"
Kirti PandyaTF or pytorch? which API seems superior?
The TWIML AI Podcast with Sam CharringtonGreat question Alisher!
Miguel Ángel García VillegasBased on your experience as an academic in linguistics, how much importance is given to machine learning in this research field nowadays?
No comments:
Post a Comment