AI and ALife as PhD themes empirical notes Luís Correia Faculdade de Ciências Universidade de Lisboa

Transcription

1 AI and ALife as PhD themes empirical notes Luís Correia Faculdade de Ciências Universidade de Lisboa Comunicação Técnica e Científica 18/11/2016

2 AI / ALife PhD talk overview PhD in general the student's point of view the supervisor's point of view PhD in AI / ALife institutional frame front line research open problems community the thesis

3 the student's side It all depends on the advisor

4 the advisor's side that's student's work Keagle Photography Library Univ Chicago

5 a compromise? Yes (the politically correct answer) depends on the advisor depends on the student depends on the institution depends on the context...

6 bottom line committing to one single cause student's motivation

7 motivated type

8 mathematical formulation Newton's 2nd law of graduation flexibility age PhD = motivation the age of a doctoral process is directly proportional to the flexibility given by the advisor and inversely proportional to the student's motivation singularity at m=0

9 the other 2 laws (for completeness sake) 1st a PhD student in procrastination tends to stay in procrastination unless an external force is applied to it 3rd for every action towards PhD there is an equal and opposite distraction

10 how to succeed? genius is 1% inspiration and 99% perspiration Thomas Edison

11 institutional integration AI is not a core subject in computer science in some institutions is regarded as marginal fundamentalists may look down on it good support from the group is important

12 AI vs. CS et al. AI has rebranded some topics of CS and other domains A* vs. Dijkstra's algorithm optimization and decision vs. Operations Research cybernetics

13 within AI hélas! fundamentalism exists also in AI areas new to AI took time to get accepted GOFAI acronym may have helped... may not be blocking but increases difficulties ALife (in silico) is marginal in AI!...

14 front line AI work deep learning

15 front line AI work autonomous cars

16 front line AI work intelligent personal assitants

17 front line AI work cognitive systems

18 front line AI work virtual (reality) worlds

19 PhD in AI AI is a scientific area requires scientific approach problem hypothesis validation

20 AI work theoretical mathematics, natural sciences prove some new theoretical results produce a new model / theory (tested with data) technique - engineering new / improved results of its application better than previous experiments supported by sound statistics

21 AI work getting fishy... AI technique applied to new type of problem framework combination of AI techniques (?) scope of MSc thesis AV O ID! methodology this is really fishy stuff... are there others to compare? does it provide an advancement in solving some problem? how to measure?

22 what is an AI thesis? original work in AI capable of synthesising into a journal paper in the end of the PhD work or after in the meantime... publish ideas in workshops publish intermediate results in conferences

23 publish or... perish

24 research report write down all your research in one single document research report it may become your PhD dissertation even if not: several papers will spin off from it

25 publishing - where? avoid scientific tourism publish in the really important venues journals and conferences generalist, or more specific ones it's harder, but better return publish in EPIA and other specific Portuguese conferences it's important to place yourself in the community

26 reference venues journals conferences Artificial Intelligence, IEEE Intelligent Systems, IEEE trans Pattern Analysis & Machine Intelligence, Data Min Knowl Discov Int'l J Comp. Vision, Med Image Analysis, IEEE T Fuzzy Sys, Int J Neural Systems, Evol Comput SIGKDD, IJCAI, AAAI, AAMAS, ECAI CVPR, ICCV, ICANN, IROS, ALife, ECAL in Portugal journal: PRIA (with Spain), AI Com (with ECCAI) conference: EPIA (biyearly) - 1st in 1985

27 interdisciplinary nature philosophy psychology linguistics neuroscience computer science & engineering ethology biology physics AI side ALife side

28 open problems in AI common sense CYC attempting... quick learning master algorithm (P. Domingos) consciousness language semantics cyborgs

29 open problems in ALife life in other support different from organic chemistry emergence of life organic & inorganic emergence of intelligence self-organisation theory of information for living systems

30 AI hot papers

31 ALife hot papers

32 awareness of other problems ethics ban develop. & use of autonomous weapons open letter signed by Hawking, Musk, Wozniak and 3,000 researchers in AI and robotics (2015) privacy issues data mining may collect & relate a lot of data

33 homework! Blade Runner Matrix (the 3 of them!) I, Robot 2001: A Space Odyssey A.I. Artificial Intelligence Ex Machina Bicentennial Man Her...

34 PhD student requirements must be able to carry independent in-depth research critical analysis capability look for additional refs. contact other researchers & motivation in the absence of these, should not continue with PhD

35 the true (motivated) PhD student defends his work! because he has built it in a solid way knowing its limitations always tries to overcome hurdles! a paper was rejected? get your act together and then... use reviews to improve your paper and resubmit it!

36 bad modelling happens...

37 PhD in the end is hardly an historical break-through Q-learning maybe the only exception in AI student should be a world class expert on his subject and he must be able to put his work in perspective

38 advisor's check-list can student be a good reviewer? can student supervise post-graduate students? would I like to have him as a colleague? would I like to have him as advisor? break the mediocrity cycle: mediocre PhD students will produce even more mediocre PhD students Michael Athans

39 some references Alan Bundy Univ. Edinburgh Manuel Bloom How to do Research at the MIT AI Lab Michael Athans, Reflections on Doctoral Research, 2000, SPDDI, UNL

40 the (untold) fun side of research

41 keep pushing!