Michael Jordan: Applied Bayesian Nonparametrics

Professor Michael Jordan

EECS Berkeley

4.30 pm, Thursday, 4 March 2010

Room G07, The Informatics Forum

10 Crichton Street

Computer Science has historically been strong on data structures and weak on inference from data, whereas Statistics has historically been weak on data structures and strong on inference from data. One way to draw on the strengths of both disciplines is to pursue the study of “inferential methods for data structures''; i.e., methods that update probability distributions on recursively-defined objects such as trees, graphs, grammars and function calls. This is accommodated in the world of ''Bayesian nonparametrics,'' where prior and posterior distributions are allowed to be general stochastic processes. Both statistical and computational considerations lead one to certain classes of stochastic processes, and these tend to have interesting connections to combinatorics.

I will focus on Bayesian nonparametric modeling based on Dirichlet processes and completely random processes, giving examples of how recursions based on these processes lead to useful models in several applied problem domains, including natural language parsing, computational vision, statistical genetics and protein structural modelling.

Embracing Uncertainty: The New Machine Intelligence

Embracing Uncertainty: The new machine intelligence

Final reminder University of Edinburgh - 18 March 2010 In this year's IET/ BCS Turing Lecture, Professor Christopher Bishop, the Chief Research Scientist at Microsoft Research Cambridge, discusses the field of machine learning, and shows how uncertainty can be modelled and quantified using probabilities. This prestigious, high profile lecture, dedicated to the memory of Alan Turing, explores how computers can 'learn'. The event is supported by the University's School of Informatics. Chris Bishop is a Professor of Computer Science at the University of Edinburgh. He is also member of the Institute for Adaptive and Neural Computation in the School of Informatics and has been elected a Fellow of the Royal Society of Edinburgh. The evening lecture will be followed by a buffet reception, offering the opportunity to meet the speaker and enjoy the surroundings of the University. To secure your place at this exciting lecture and buffet reception. The lecture is free of charge but there is a charge should you wish to attend the buffet reception. The lecture and buffet reception are both open to members and non-members of BCS and the IET. This year, booking for the event will be via the IET website and BCS Members will need to register an account on the website before being able to book on to the lecture.

Winter Wonderland: ww.inf.ed.ac.uk

Today we had a snowball fight. See picasa for more images of the roof garden in the snow.

Creating Knowledge in the Age of Digital Information

A Public Lecture

Bob Constable, Cornell,
SICSA Distinguished Visiting Fellow

10th November 2009 17:10-18:00

Lecture Theatre 5
Appleton Tower
11 Crichton Street
Edinburgh

This talk will illustrate a fundamental feature of informatics that explains its relevance to every academic discipline, namely the role of computing and digital information technologies and ideas in accelerating discoveries and creating knowledge totally inaccessible without them.

The impact of informatics on the academy will be large, changing how we create, preserve and disseminate knowledge - that's the core university mission. This centrality to the university's mission increases demand for education in this field that serves the needs of all students. It requires us to teach broad computational thinking and create a basis for life-long learning in this rapidly changing field.

This lecture is supported by SICSA and IDEA lab Edinburgh.

Professor R. Wayne Davies

Welcome to Wayne Davies who joins us as an Honorary Professor.
Wayne has a unique combination of international level scientific achievement with extensive business experience in the field of biotechnology. After studying medicine at Cambridge he was caught up in the molecular biology revolution and switched to pure science, working with Sydney Brenner in Cambridge, Bill Dove in Madison and Benno Muller-Hill in Cologne before going solo. Work on gene regulation and integration of lambda bacteriophage, including publication of the integration site sequence, was followed by a move into eukaryotes and a series of key contributions to molecular biology of filamentous fungal systems, including the first transfection method and mitochondrial genome structure. His group discovered fungal self-splicing introns, and he published the first secondary structure of a ribozyme, and the internal guide sequence model for splice-site selection, which also explained how the Tetrahymena intron (for the discovery of which Tom Cech got a Nobel) actually worked. For this work he was elected a member of EMBO, and then moved into the biotech industry as Vice-President for Science of Canada’s then flagship biotechnology company, Allelix Biopharmaceuticals, where he managed a staff of 70 scientists for six years. Academic work since returning to the UK to become Professor of Biotechnology at the University of Glasgow has focussed on molecular neuroscience (neurodegeneration mechanisms, CNS myelin studies) and more recently on tissue ageing and the discovery with Dr. Paul Shiels of a new type of repair stimulator cell with real therapeutic potential for diabetes and other organ-damage diseases.
Wayne has increasingly focussed on applied science, licensing opportunities and spin-out companies. With Kim Kaiser and Alun Davies he spun out Neuropa Ltd. and ran that company as CEO for 5 years. He was then seconded as CEO to UmanGenomics AB of Umeå, Sweden for two years. In these roles he had hands-on experience of biotechnology company management, Board membership, multi-round fundraising from venture capital and local Angel sources and interaction with major pharmaceutical companies. He recently found US investors to form Pathfinder LLC to exploit the potential of repair-stimulator cells.
His involvement with the University of Edinburgh and Informatics began when he joined with Douglas Armstrong to develop the Brainwave project into a commercial venture. This will result in the near future in the spinning out of Brainwave-Discovery Ltd., a company that provides custom assay development services for the pharmaceutical industry and will develop novel technology platforms for CNS drug discovery. As a result, Wayne expects to spend more time in Edinburgh, and is looking forward to providing any help to colleagues that is congruent with his experience, whether in mammalian neuroscience and stem cell biology or in company formation and development.

For Alan Turing, a real apology for once

I had intended to mark this news with a post of my own. However, Geoff Pullum does it so much better that, with his permission, I just reproduce here Geoff's Language Log post.
In an age where many purported public apologies are just mealy-mouthed expressions of regret ("I'm sorry it all happened"), or grudging self-exculpatory conditionals ("If some people think I shouldn't have said it, I'm sorry they were upset"), it is good to see a genuine and direct apology for once, addressed (though more than half a century too late) to a man who deserved admiration, gratitude, and respect, but was instead hounded to death. The UK Prime Minister, Gordon Brown, has released a statement regarding the treatment of Alan Turing in the early 1950s, and the operative words are:

on behalf of the British government, and all those who live freely thanks to Alan's work I am very proud to say: we're sorry, you deserved so much better.

That's how to say it (ignoring the punctuation error — the missing comma after work): not a bunch of evasive mumbling about how unfortunate it all was, but a simple "We're sorry."

Turing did indeed deserve so much better. He created modern theoretical computer science; opened fundamental new areas of mathematical logic; made very important contributions to other areas of mathematics (e.g., the technique known as Good-Turing frequency estimation in statistics); and most importantly, he gave up his academic work during the Second World War to work at Bletchley Park on the extremely difficult task of decrypting German communications encrypted with the Enigma machine. The Bletchley Park team did succeed, and thus the Royal Navy became able to read the content of all the Nazis' messages to U-boats in the North Atlantic. It was a crucial turning point in the war. But a mere seven years later, a young man shared Turing's bed for the night in Manchester, and later helped someone burgle the house, and Turing naively reported the theft to the police. The police reaction was to arrest Turing, because they guessed what had been going on. "Gross indecency" was the charge (it is the British legal euphemism for cocksucking). Turing had a choice between serving prison time or agreeing to chemical castration, a medicalized "cure" for his presumed abnormality. He bore the latter for two years and then took cyanide. The way British mid-20th-century sex law drove him to suicide was genuinely something for the country to be ashamed of. It was good to see the official apology (which hundreds of eminent scientists had asked the Prime Minister to express).

September 11, 2009 @ 4:47 am · Filed by Geoffrey K. Pullum

Professor Keith van Rijsbergen

Welcome to Keith van Rijsbergen who joins us as an Honorary Professor.
Keith van Rijsbergen was born in Holland in 1943. He was educated in Holland, Indonesia, Namibia and Australia. He took a degree in mathematics at the University of Western Australia. As a graduate he spent two years tutoring in mathematics while studying computer science. In 1972 he completed a Ph.D. in computer science at Cambridge University. After almost three years of lecturing in information retrieval and artificial intelligence at Monash University he returned to the Cambridge Computer Laboratory to hold a Royal Society Information Research Fellowship. In 1980 he was appointed to the chair of computer science at University College Dublin; from there he moved in 1986 to Glasgow University.
Since about 1969 his research has been devoted to information retrieval, covering both theoretical and experimental aspects. He has specified several theoretical models for IR and seen some of them from the specification and prototype stage through to production. His current research is concerned with the design of appropriate logics to model the flow of information. He has been involved in a number of EU projects and working groups on IR, including Fermi, Miro, Mira, Idomeneus, and more recently K‐space. He is a fellow of the Royal Academy of Engineering, Royal Society of Edinburgh, IEE, BCS, and ACM. In 1993 he was appointed Editor‐in‐Chief of The Computer Journal, an appointment he held until 2000. He has served as a programme committee member and editorial board member of the major IR conferences and journals. He is the author of a well‐known book Information Retrieval, Butterworths, 1979. In 1999, together with Crestani and Lalmas,he published a book entitled "Information Retrieval: Uncertainty and Logics". His most recent book is The Geometry of Information Retrieval, CUP, 2004. Some of his research papers can be accessed at http://www.dcs.gla.ac.uk/~keith/. He was chairman of the most recent REA panel in Computer Science and Informatics. He is chairman of the panel for Computer Science and Informatics (PE6) for the European Research Council.

Congratulations to John Lee

Congratulations on the Chancellor's Award for Teaching

HRH The Prince Philip Duke of Edinburgh presented the Chancellor's Award for Teaching to John Lee at a gala dinner in the Palace of Holyroodhouse.

John Lee has led development of a system known as "YouTute", supported by a grant from the Principal's e-Learning Fund (2007-8). In this system, video recordings of tutorial discussions are made available to students. The recordings of entire tutorials (from two cameras and a Smartboard) are streamed in an online environment in which students can extract "virtual clips" from the videos, annotate, tag and comment these, share them, and keep them for future reference. The videos are presented accompanied by the tutorial question sheets and, where appropriate, solutions, as well as the relevant lecture slides. This creates a new kind of learning resource, around which the students can develop a collaborative learning activity that will promote reflection, deepen understanding, and add significantly to the value of the original tutorial experience. Students can also observe different approaches to particular problems (and to teaching and learning) as revealed in different tutorial groups' discussions.

A robust prototype system is currently being trialled in Informatics, with great potential also for use in many other areas of the University. Over 50 tutorials were recorded in AY 2007-8 from Informatics 2A/B,which have been made available to all second year Informatics students since then (nearly 400 students). The system was especially appreciated by resit students revising during the summer, who used the system for over 50 hours, suggesting that it may be very helpful for the slower learners.

Collection of tutorials continues. The next step is to integrate recordings of lectures, which are increasingly captured in many subject areas, allowing these also to be re-used in a much more flexible and substantial way. The system is also the focus of research to assist students further by developing automated means of indexing and tagging the videos, exploiting methods developed in existing Informatics research on multi-party meetings.

YouTute exploits vicarious learning, which is learning from exposure to the learning experiences of others. John Lee has a long history of working to develop innovative ways of exploiting the concept of vicarious learning. He has been investigating this idea, with collaborators, in a series of projects since 1995, funded by EPSRC, ESRC, and the Teaching and Learning Research Programme — the recent Principal's e-Learning Fund grant has enabled a sharper focus on application to be added to this work. There is clear evidence that vicarious learning has substantial benefits for motivation and attitude as well as discussion skills and learning strategies. Research elsewhere also suggests that vicarious learners learn better if they collaborate with each other. It is to exploit these benefits, and also to evaluate subject-specific learning in much more detail, that practical application has become the focus, leading to the current YouTute deployment.

John has been keen to promote innovations in teaching for many years. 10 years ago, in what is now the School of Arts, Culture and Environment, he co-developed the highly successful MSc in Design and Digital Media. He introduced teaching of web design and technologies at a very early stage, and continues to teach these and direct this programme. In his seconded position in Informatics, he was the founding course organiser of the pioneering Informatics Entrepreneurship courses, one of which has also been adapted for use in Design and Digital Media.

Vijay Nagarajan

I am delighted to announce the appointment of Vijayanand Nagarajan to a lectureship in Informatics, funded by the Numerical Algorithms and Intelligent Software Centre.

Vijay 's research interests lie in the areas of compilers, computer architecture and software engineering. He plans to work within ICSA on problems that span these areas. He plays cricket, violin and electric guitar.

Vijay received his MS degree in Computer Science from the University of Arizona in 2005. He spent the summer of 2006 as a research intern at the Intel Programming Systems Laboratory, and has studied for his PhD under the direction of Prof. Rajiv Gupta, in the Department of Computer Science and Engineering at the University of California, Riverside. His dissertation proposes an efficient and programmable runtime monitoring approach for multicores, which can be used to increase the performance and reliability of parallel programs running on such architectures.

Vijay expects to defend his PhD in August 2009, and will then join the school in October.

Guido Sanguinetti

I am delighted to announce that Dr Guido Sanguinetti will join the School of Informatics on 1st August, as a SICSA lecturer.

Guido received his MSc (Laurea) in Physics in his native Genova, and then pursued a D.Phil. in Mathematics at the University of Oxford, working on algebraic and differential geometric methods to solve nonlinear differential equations.

After two years as a professional musician, he returned to research in 2004 as a postdoc and then a lecturer in Computer Science in Sheffield. His main research interests lie in reverse engineering dynamical systems governed by sets of (stochastic) differential equations, with a particular focus on applications in systems biology.

Guido will take a leave of absence to allow him to complete some projects at Sheffield, but he plans to make several extended visits to Edinburgh before moving to Scotland in 2010.

Phil Wadler — SIGPLAN Chair

Congratulations to "Phil Wadler on his election as Chair of SIGPLAN 2009-2012

Phil says, "Programming languages stand poised to take centre stage, as the web and multicores push distributed and concurrent computing to the forefront."

Rod Burstall: Programming Languages Achievement Award

Congratulations to Rod Burstall

Rod has made deep, seminal contributions to the design of programming languages and the field of program verification. These contributions, which many of us now take for granted, include the introduction of algebraic datatypes coupled with pattern-matching clausal function definitions as found in Hope, ML, Haskell and Coq; the generalization and use of structural induction for proving properties of programs; the fold-unfold method for deriving efficient, provably-correct programs from easy to understand prototypes; mechanisms for reasoning about pointer-based, imperative programs that directly led to the development of separation logic; proof techniques and connections to modal logic for reasoning about concurrent programs; and the use of dependent types and algebraic specifications for constructing module systems that directly influenced SML and OCaml. Through these amazing contributions and his collaborations and mentorship, he helped build one of the most important centers of programming research at Edinburgh, which was eventually institutionalized as the Laboratory for Foundations of Computer Science.

Amin Coja-Oghlan: Best Paper Award ICALP 2009

Congratulations to Amin Coja-Oghlan

Amin's paper, A better algorithm for random k-SAT will receive the Best Paper Award of ICALP 2009, Track A.

The award will be conferred at the EATCS General Assembly on Tuesday, July 7

Charles Sutton

I am delighted to announce the appointment of Dr. Charles Sutton to a SICSA lecturership in the School of Informatics.

Charles has been a postdoctoral researcher at the University of California, Berkeley, since 2007. His recent work has aimed at new statistical machine learning methods designed to aid the management of large-scale computer systems. In particular, he has developed methods for performance modeling that are rooted in machine learning, applying them to the control, visualization, and diagnosis of distributed Web applications. More generally, his research interests include machine learning, graphical models, approximate inference, structured prediction, natural language processing, and the application of machine learning methods to computer systems. Charles received his PhD from the University of Massachusetts Amherst in 2008. His thesis work concerned efficient training methods for conditional random fields, with applications in natural language processing.

Iain Murray

I am delighted to announce that Dr Iain Murray will join the School of Informatics on 1st August, as a SICSA lecturer and a member of the Institute of Adaptive and Neural Computation.

Iain received MA and MSci degrees in Natural Sciences (Physics) from the University of Cambridge before obtaining a PhD from the Gatsby Computational Neuroscience Unit at University College London. His thesis introduced a range of new 'Markov chain Monte Carlo' algorithms for solving integrals in hard statistical inference problems. While at Gatsby Iain also developed strong interests in probabilistic modelling and efficient algorithms for solving inference problems.

Partly supported by a Canadian Commonwealth Research Fellowship, Iain moved to Toronto in 2007 and joined the Machine Learning group there as a postdoctoral fellow. He has continued to expand the applicability of Markov chain Monte Carlo methods for statistical applications, such as the evaluation of large-scale probabilistic models. Iain has also formed collaborations to apply and extend hierarchical Bayesian methods. Recent application areas include understanding human perception and inferring celestial dynamics.

Iain will take a leave of absence to allow him to complete his Fellowship in Toronto, but he plans to make several extended visits to Edinburgh before moving to Scotland in 2010.

Third FRS for Informatics@Edinburgh — but only one Oscar.

Congratulations to Peter Buneman on his election as a Fellow of the Royal Society.

A total of eleven past and present distinguished Edinburgh informaticians have been elected to this fellowship. Of these, Peter Buneman, Robin Milner and Gordon Plotkin are current members of the School.

Peter Buneman is distinguished for his advances in uniting programming languages and databases. On the theoretical side this has involved new results in types, monads and structural recursion including (with his student Ohori) type inference for record types, and (with Tannen et al) results that demonstrated a tight connection between monad-based languages and those based on the predicate calculus. On the application side, he used these techniques to demonstrate that -- contrary to an assertion by the US Department of Energy -- queries on existing non-relational genomic databases could be directly evaluated; fruitful collaboration with biologists ensued.

This research on databases and languages carries over into his recent study of the principles of semistructured or "web-like" data of which he is a leading proponent, and co-author of the first text book in this new field. Another recent concern is with the provenance of data on the Web, where data is continually copied and transformed. Already, with Khanna et al. he has built an efficient archiving system for scientific databases; more fundamentally, he seeks a formal basis for tracing provenance.

In addition to his work in databases, Buneman's early work on mathematical phylogeny underlies most modern phylogenetic reconstruction techniques.

The Space and Motion of Communicating Agents

I am delighted to announce that Robin Milner is returning to the University of Edinburgh, part-time. He will be a SICSA Advanced Research Fellow and will hold the Chair of Computer Science.

Robin Milner graduated from Cambridge in 1958. After short posts he joined the University of Edinburgh in 1973, where he co-founded the Laboratory for Foundation of Computer Science in 1986. He was elected Fellow of the Royal Society in 1988, and in 1991 won the ACM's A.M. Turing Award. He joined Cambridge University in 1995, headed the Computer Laboratory there for four years, and retired in 2001. His research achievements (often joint) include: the system LCF, a model for many later systems for interactive reasoning; Standard ML, an industry-scale but rigorously based programming language; the Calculus of Communicating Systems (CCS); the Pi Calculus.

Currently he works on Bigraphs, a topographical model which aims to provide a theoretical foundation for mobile interactive systems. Cambridge University Press has just published The Space and Motion of Communicating Agents, Robin's book on this area.

Robin will be giving a short course on bigraphs in Edinburgh on May 13th and 14th 2009. So that we can keep track of numbers, please sign up for the course at http://milner-bigraphs.eventbrite.com (password is milner).

The Forum appears in Google Earth

Google has updated Google Earth with new images for Edinburgh Southside, dated 21 July 2008.

Open the title link with Google Earth to see the George Square masterplan superimposed on the Forum.

Informatics welcomes Peter Higgs

Peter Higgs was the first student of Christopher Longuet-Higgins, one of the forefathers of Informatics at Edinburgh.

The School is delighted to host Peter's recent portrait by Ken Currie. The portrait will hang in the Forum for the next few months, until it moves to its permanent home in the James Clerk Maxwell Building.

Mark Steedman and David Willshaw were also students of Longuet-Higgins.

RAE 2008 results: Informatics and Computer Science

HEFCE has published preliminary results of the RAE.

These comprise tabulated overall quality profiles per unit of assessment (UoA) and per institution. Each quality profile presents the proportions, rounded to 5 per cent, of research activity in each submission judged to have met each of the quality levels from 4* to unclassified. Alongside the quality profile for each submission, HEFCE publish the full-time equivalent (FTE) number of Category A staff included in the submission.

The highest, 4* grade is awarded for work that is world-leading; 3* is for internationally excellent, and 2* for internationally recognised research.

4*

Quality that is world-leading in terms of originality, significance, and rigour

3*

Quality that is internationally excellent in terms of originality, significance, and rigour but which nonetheless falls short of the highest standards of excellence

2*

Quality that is recognised internationally in terms of originality, significance, and rigour

1*

Quality that is recognised nationally in terms of originiality, significance, and rigour

U/C

Quality that falls below the standard of nationally recognised work, or which does not meet the published definition of research for the purposes of this assessment

Taking the profile percentages of the FTE staff number gives a research volume profile for each submission.

H-0167 University of Edinburgh
UoA 23 Computer Science and Informatics

4* volume

36.4875

3* volume

52.125

2* volume

15.6375

total:

104.25 FTE

These results confirm that the University of Edinburgh is, once again, the strongest in the UK for research in Informatics and Computer Science.

Informatics at Edinburgh delivers more world-leading (4*) research than anyone else — 69% more than our nearest competitor.

We contribute 10% of the UK's world-leading research in this UoA.

We deliver more internationally-excellent (3*) or world-leading (4*) research than anyone else — 44% more than our nearest competitor. Edinburgh contributes more research that is at least internationally excellent than Oxbridge.

Overall, ours is the largest research grouping in the UK — again, 44% larger than our nearest competitor.

All of our research is internationally recognised, or better (2*, 3* or 4*) — and we submitted every eligible member of staff.

---

Ranking

The use of grade profiles means that there is no single ranking as a result of this RAE.

The detailed formulae for research funding based on the RAE will not be announced by funding councils for several months. However, future research funding is expected to be based on a weighted sum over the RAE volume profile.

For any score given as a weighted sum S = α 4* + β 3* + γ 2* + δ 1*, where &alpha > 0 and α ≥ β ≥ γ ≥ δ ≥ 0, Edinburgh scores top in the UK.

---

Progress

Our volume of world-leading and internationally excellent research (88.6 FTE) exceeds our entire submission to RAE 2001 (80.2 FTE). In addition, we have recently made half a dozen outstanding new appointments, not included in this assessment, that further increase our research excellence and capacity. We have also recently improved our environment by moving into the Informatics Forum — acclaimed as the best building in Scotland. This building, designed for interaction, brings an eclectic multidisciplinary collection of over 500 researchers in Informatics and Computer Science together, under one roof.

---

Collaboration with Linguistics

Edinburgh has the UK's strongest research group in Linguistics, with more world-class (4*) research in Linguistics (10.8 FTE) than any two other submissions combined. Through the Human Communication Research Centre (HCRC) and the Centre for Speech Technology Research (CSTR), we work closely with these colleagues in the areas of Speech and Language Processing, Computational Linguistics and Cognitive Science.

---

Informatics and Computer Science in Scotland

The Scottish Informatics and Computer Science Alliance (SICSA) brings together internationally excellent research from across Scotland, to double the University of Edinburgh's 3* + 4* volume. This critical mass of excellence makes Scotland a match for any leading centre of Informatics and Computer Science, worldwide.

SICSA is actively recruiting, and plans to complete appointments to some 30 new posts by the end of 2010. Our policy is to recruit internationally, to make outstanding appointments that raise our research quality, and further strengthen Scotland's profile in Informatics and Computer Science.

---

Informatics and Computer Science across the UK

Eighty-one submissions were made to UoA 23. The top ten (by 4* volume) account for over 50% of the UK's world-leading research in this UoA.

Manchester is our nearest competitor. Cambridge, Oxford, University College, and Imperial College — almost indistinguishable from one another — come next, followed by Southampton; Bristol, Birmingham and Nottingham complete the top ten.

---

The Graph

The abscissa is a parameter β

The ordinate (plotted on log scale) is V = 4* + β 3*

The point is that whatever reasonable combination of 4* and 3* volume you choose to take, Edinburgh comes out tops.

The use of a log scale makes it easier to compare relative sizes. For example, the gap between Cambridge and Manchester at the right-hand end of the graph signifies that Manchester has roughly 50% more 3* + 4* volume than Cambridge. A slightly larger gap at the left-hand end between Manchester and Edinburgh signifies that Edinburgh has more than 50% more 4* volume than Manchester.