Weekly QuEST Discussion Topics, 31 Aug

QuEST 31 Aug 2018

We will have a short discussion this week to allow people to get an early start on Labor Day weekend.  We want to start by revisiting the Communications topic – specifically critiquing the Vision / Steps / Contributions for ACT3.

We also want to comment on two email threads from this week.  The first is in some recent Question-Answering systems work.

Deep contextualized word representations
Matthew E. Peters et al – from the Allen AI Institute

Proceedings of NAACL-HLT 2018, pages 2227–2237

New Orleans, Louisiana, June 1 – 6, 2018. c 2018 Association for Computational Linguistics

• We introduce a new type of deep contextualized word representation that models both
• (1) complex characteristics of word use (e.g., syntax and semantics), and
• (2) how these uses vary across linguistic contexts (i.e., to model polysemy – the coexistence of many possible meanings for a word or phrase).
• Our word vectors are learned functions of the internal states of a deep bidirectional language model (biLM), which is pretrained on a large text corpus.
• We show that these representations can be easily added to existing models and significantly improve the state of the art across six challenging NLP problems, including question answering, textual entailment and sentiment analysis.
• We also present an analysis showing that exposing the deep internals of the pre-trained network is crucial, allowing downstream models to mix different types of semi-supervision signals.

Questioning and Answering in context also

arXiv:1808.07036v2 [cs.CL] 22 Aug 2018

QuAC : Question Answering in Context

We present QuAC, a dataset for Question Answering in Context that contains 14K information-seeking QA dialogs (100K questions in total). The interactions involve two crowd workers: (1) a student who poses a sequence of freeform questions to learn as much as possible about a hidden Wikipedia text, and (2) a teacher who answers the questions by providing short excerpts from the text. QuAC introduces challenges not found in existing machine comprehension datasets: its questions are often more open-ended, unanswerable, or only meaningful within the dialog context, as we show in a detailed qualitative evaluation. We also report results for a number of reference models, including a recently state-of-the-art reading comprehension architecture extended to model dialog context.  Our best model underperforms humans by 20 F1, suggesting that there is significant room for future work on this data. Dataset, baseline, and leaderboard are available at http://quac.ai.

The second thread provides some great historical context

Cognitive Systems Research 50 (2018) 83–145

From cybernetics to brain theory, and more: A memoir

Michael A. Arbib

While structured as an autobiography, this memoir exemplifies ways in which classic contributions to cybernetics (e.g., by Wiener,

McCulloch & Pitts, and von Neumann) have fed into a diversity of current research areas, including the mathematical theory of systems and computation, artificial intelligence and robotics, computational neuroscience, linguistics, and cognitive science. The challenges of brain theory receive special emphasis. Action-oriented perception and schema theory complement neural network modeling in analyzing cerebral cortex, cerebellum, hippocampus, and basal ganglia. Comparative studies of frog, rat, monkey, ape and human not only deepen insights into the human brain but also ground an EvoDevoSocio view of “how the brain got language.” The rapprochement between neuroscience and architecture provides a recent challenge. The essay also assesses some of the social and theological implications of this broad perspective.

Weekly QuEST Discussion Topics, 24 Aug

QuEST 24 Aug 2018

We will start by just mentioning a thread associated with the book “The Righteous Mind – why good people are divided by politics and religion” by Jonathan Haidt. 

Really interesting book overall, but one really cool thought was his theory about the ‘why’ behind the evolution of sys 2.  His idea is that sys 2 serves to create post hoc explanations of the decision making process, for the main purpose of defending the result of our intuition’s judgement to ourselves and to others.  Without this piece, humans as a species would be unable to influence each other’s behavior, to create trust and cohesion amongst groups (tribes, villages, cities, nations.)

The result of reasoning only has a very weak impact on my own intuition. We rarely seek out evidence on our own that runs counter to our current beliefs.  Thinking/reasoning is confirmatory rather than exploratory.  It’s main impact lies in explaining things to another agent to pull them to my cause.  

Humans create rationales they believe account for their judgments, but they are post hoc.  Intuition launches reasoning, but does not depend on the success or failure of that reasoning 

You can’t change people’s minds by refuting their arguments (my reasoning to your reasoning).  Have to impact their intuitive process.  

 

For the remaining time this week’s QuEST meeting will be a workshop focused on material recently provided to us by our colleague Prof Winston of MIT.  Specifically, on many occasions we have witnessed our inability to communicate what we do.  Prof Winston gives a lecture series at MIT on “Communication”.  He is writing a book to capture those ideas and they are extremely useful for all of us to absorb.  Whether you are presenting to your colleagues in your branch / division / directorate or the Scientific Advisory Board or at a conference or you are putting together an ‘investor deck’ or ‘tear sheet’ there are universals on lessons we should all learn from to make you an effective communicator!  These lessons will help you speak / write / instruct / interview …

There is far too much material to cover on this topic but I needed to make the group aware this is something you can get better at and we need each other to have this ability.  For the purpose of this QuEST meeting I want people to come prepared with your ‘story’.  What do you do?  To prepare to tell the group what you do I want you to construct your personal story.  Prof Winston suggest the V-S-N…C framework (vision, steps, news, … exposition, contributions). 

A common mistake made by many is NOT to provide the most important part of your communication soon enough.  You have to capture the listener / reader immediately.  You should start with a quick clear statement of your vision.  The vision has to explain what you are doing is associated with a problem that the listener cares about and an approach to solving that problem.  When we talk to people about what we do we too often focus on the technology and not the problem.  I have had to advise many SAB presenters to explain the problem this work will lead to solutions for.  In my company if I was talking to the board of directors the talk would present technology from the perspective of the problem of revenue / costs not technology advances by themselves.  Too often presentations to VCs focus on technology and NOT what they care about which is a new way for them to make money.  Know your audience and tailor your Vision to be the ‘problem’ they will care about.  For the purposes of this QuEST meeting assume your audience are visitors to your organization that are tax payers and citizens.  This part of the material may take the whole time we have – I want people to provide their vision statement and others comment on ideas.  There are a whole set of lessons associated with how to be critiqued.  So we will fold those in during this first discussion also.  One of the lessons we will violate for the purpose of QuEST but in general you want to see comments sequentially.  Otherwise a group of reviewers will focus on the same issues – the obvious ones.  You really want the critiques to come from people who will be brutally honest and the criticisms to be actionable. 

Bottom line – we want to use the brilliant insights of Prof Winston to help the QuEST  to become better communicators.

 

 

 

Weekly QuEST Discussion Topics, 10 Aug

QuEST 10 Aug 2018

We want to focus this week on multi-agent systems.  Specifically our colleague Prof Bert had sent me some articles that I finally worked through while on the road this week and I want to have a discussion on our Knowledge Platform (KiP).

The first article:

Agents and Service-

Oriented Computing

for Autonomic Computing

A Research Agenda

Frances M.T. Brazier • Vrije Universiteit Amsterdam

Jeffrey O. Kephart • IBM T.J. Watson Research Center

1. Van Dyke Parunak •Tech Team Government Solutions

Michael N. Huhns • University of South Carolina

Autonomic computing is the solution proposed to cope with the complexity of today’s computing environments. Self-management, an important element of autonomic computing, is also characteristic of single and multiagent systems, as well as systems based on service-oriented architectures. Combining these technologies can be profitable for all — in particular, for the development of autonomic computing systems.

The second article:

Software engineering for self-organizing systems

1. VAN DYKE PARUNAK1and SVEN A. BRUECKNER2

1AxonConnected, 2322 Blue Stone Hills Drive, Suite 20, Harrisonburg, VA 22801;

e-mail: van.parunak@axonconnected.com;

2AxonAI, 2322 Blue Stone Hills Drive, Suite 20, Harrisonburg, VA 22801;

e-mail: sven.brueckner@axonai.com

Self-organizing software systems are an increasingly attractive approach to highly distributed, decentralized, dynamic applications. In some domains (such as the Internet), the interaction of originally independent systems yields a self-organizing system de facto, and engineers must take these characteristics into account to manage them. This review surveys current work in this field and outlines its main themes,

identifies challenges for future research, and addresses the continuity between software engineering in

general and techniques appropriate for self-organizing systems.

Weekly QuEST Discussion Topics, 3 Aug

QuEST 3 Aug 2018

We want to start this week with a thread that was initiated by our colleague Mike Y – on the Generative Query Network material from DeepMInd.

To Make Sense of the Present,
Brains May Predict the Future
By J O R D A N A C E P E L E W I C Z

https://www.quantamagazine.org/to-make-sense-of-the-present-brains-may-predict-the-future-20180710/

A controversial theory suggests that perception, motor control, memory and other brain functions all depend on comparisons between ongoing actual experiences and the brain’s modeled expectations.

QuEST has suggested in fact the conscious perception is in fact an expectation modulated by the sensed data – a confabulation, an ‘imagined present’ in the words of Edelman

Some neuroscientists favor a predictive coding explanation for how the brain works, in which perception may be thought of as a “controlled hallucination.”

This theory emphasizes the brain’s expectations and predictions about reality rather than the direct sensory evidence that the brain receives.

Last month, the artificial intelligence company DeepMind introduced new software that can take a single image of a few objects in a virtual room and, without human guidance, infer what the three-dimensional scene looks like from entirely new vantage points.

Given just a handful of such pictures, the system, dubbed the Generative Query Network, or GQN, can successfully model the layout of a simple, video game-style maze.