Archive for July, 2013

Weekly QUEST Discussion Topics, July 31

QUEST Discussion Topics and News
July 31, 2013

Two main discussion points for today’s meeting

The first is sensors versus sensing – I would appreciate a couple of minutes to work through a series of talking points / story board for convincing people we need to transition from a sensors centric view of development to a sensing one

The second topic is QUEST for Cognitive Electronic Warfare – this is a revisit to the material that Prof Mills provided us last year – but in preparation for an upcoming visit I have to Oak Ridge National Lab I want us together to brainstorm on how what we discuss in QUEST can be applied to the modern issues we face in EW.


Weekly QUEST Discussion Topics and News, July 24

1.) We will start this week by answering any questions people had about the ‘Challenges and Opportunities with Big Data’ article by Agrawal et al. – this was the last topic I brought up as we closed last week and several people wanted copies of the article – it will be posted on the VDL and the blog cite. The idea I was pushing is to ‘rethink data analysis in fundamental ways’ – I’m concerned about rethinking the DCGS (distributed common ground station) problem but have been talking to other government and commercial entities that have similar big data / human bandwidth issues. I also was tying this discussion to my fascination with the potential for crowd sourcing the knowledge generation aspects of much of our data processing and I think it also has great potential in commercial areas.
2.) The second topic is the answer remaining questions about ‘meaning’ – recall from last week: Walkaway definition of meaning: maybe I should have said the meaning of a piece of data to a given agent (recall it can’t be termed data without specifying the agent that is consuming it) is all the information that is generated by that given Agent as a result of processing that piece of data into an infon (piece of information) — all the links / qualia — that is information may begot information since it can become data to subsequent processing – a given piece of data that gets converted to information by an agent can use that information as data in subsequent processing to generate additional information – the sum total of the impact to the representation of an agent by a piece of data is what the piece of data ‘means’ to that Agent
3.) That discussion on meaning brought up our position on Gists/Links – so the last topic for this week is a discussion on Gists – Links
4.) The last topic that took some bandwidth this week for me was an article in this month’s Sci American on ‘to sleep perchance to prune’ – on sleep and memory – so if time allows I will provide a synopsis on that article.

Weekly QUEST Discussion Topics and News July 24

Big Data white paper

Categories: News Stories

QUEST Discussion Topics and News, July 17

1.) The first topic this week will be an introduction to Conditional Random Fields – the goal of this discussion is to allow our math group to be exposed to an ‘extension’ of the Hidden Markov Model work they’ve been interested in – CRFs condition on the observations without modeling it thus avoiding independence assumptions and can accommodate long-range dependencies at multiple scales (statistical dependencies exists between the entities we wish to model – they are situated) – our experience with this approach was led by our colleague Bobby Bireer during his investigation of classifying web documents and we needed a means to capture the importance of links that defined the relationships between pages in addition to the text content. This comes up as their use in natural language processing (which of course we have been pursuing as language is the means critters communicate aspects of their internal conscious representation) has been impressive.
2.) The second topic is to continue our discussion on ‘meaning’. The goal of code breaking (to include our analogy of code breaking to the Turing test) – is to come up with a representation of the ‘meaning’ of a set of symbols to the transmitter of those symbols – and that meaning is deduced by aligning the transmitted symbols with a set of symbols that are in the possession of the code breaker – (when humans have normal conversations this occurs – the speaker uses their qualia {internal symbols available to the speaker} to generate a set of words to utter {symbols the speaker can use to communicate with another person} – that are received by the person that is being talked to, those words that are heard {symbols by the person who is listening} and then those are transformed into qualia by the listener {internal symbols in the mind of the listener}.) the key in the conversation is for the listener’s qualia symbol set to get aligned with the speaker’s – but as in all code breaking it requires a sequence of alignments being accomplished. The hope of the speaker is the evoked qualia in the mind of the listener have the intended meaning to that listener. This leads to the need to define ‘meaning’.
3.) The other topics are on list to discuss include a discussion of the potential use of crowdsourcing to handle many of our manning issues associated with analyst generating intelligence products. ‘are we using our analyst incorrectly?’ and also we had some discussion that we needed to finish on equations capturing intelligence amplification for the specific example of breast cancer possibly extending to our DCGS issues.

Weekly QUEST Discussion Topics and News 13 July

Bio for Dr. Divakaran and an abstract for his presentation to QUEST this week

Title: Multimodal analysis of human behavior for social interaction training.


We present a suite of multimodal techniques for assessment of human behavior with cameras and microphones. These techniques drive the sensing module of an interactive simulation trainer in which the trainee has lifelike interaction with a virtual character so as to learn social interaction. We recognize facial expressions, gaze behaviors, gestures, postures, speech and paralinguistics in real-time and transmit the results to the simulation environment which reacts to the trainee’s behavior in a manner that serves the overall pedagogical purpose. We will describe the techniques developed and results, comparable to the state of the art ,obtained for each of the behavioral cues, as well as identify avenues for further research. Behavior sensing in social interactions poses a few key challenges for each of the cues including the large number of possible behaviors, the high variability in execution of the same behavior within and across individuals and real-time execution. Furthermore, we have the challenge of appropriate fusion of the multimodal cues so as to arrive at a comprehensive assessment of the behavior at multiple time scales We also explore going beyond extraction of verbs to extraction of the associated adverbs that capture the intensity/valence of the behavior.. We will present a video of the demonstration of the end to end simulation trainer. If time permits, we will cover other training applications such intelligent tutoring systems and stress resiliency training systems.


Ajay Divakaran, PhD is a Technical Manager at SRI International Sarnoff. He has developed several innovative technologies for multimodal systems for both commercial and government programs over the past 16 years. He currently leads SRI Sarnoff’s projects on Modeling and Analysis of Human Behavior for the DARPA SSIM project, ONR Stress Resiliency project, Army “Master Trainer” Intelligent Tutoring project, Audio Analysis for Event Detection in Open Source Video in the IARPA Aladdin program, and People, Vehicle and Vessel tracking for ONR and JIEDDO-DHS among others. He worked at Mitsubishi Electric Research Labs for ten years where he was the lead inventor of the world’s first sports highlights playback enabled DVR, as well as a manager overseeing a wide variety of product applications of machine learning. He was elevated to Fellow of the IEEE in 2011 for his contributions to multimedia content analysis. He developed techniques for recognition of agitated speech for his work on sports highlights. He established a sound experimental and theoretical framework for human perception of action in video sequences, as lead-inventor of the MPEG-7 video standard motion activity descriptor. He serves on TPC’s of key multimedia conferences and served as an associate editor of the IEEE transactions on Multimedia from 2007 to 2011 and has two books and over 100 publications to his credit as well as over 40 issued patents. He received his Ph.D. degree in Electrical Engineering from Rensselaer Polytechnic Institute in 1993.



We will NOT be meeting this week for QUEST due to the holiday. We will resume our meetings next week, on Wednesday July 10th. Also, a reminder that the weekly meetings will regularly be on Wednesdays due to the shortened work week going forward.

Have a safe and happy 4th of July everyone.