#] #] ********************* #] "$d_webRawe"'My sports & clubs/SERC systems engng/0_SERC notes.txt' www.BillHowell.ca 24Dec2020 initial To view this file - use a text editor (not word processor) constant width font (eg courrier 10), tab - 3 spaces I found out about this, I think, through the Stevens Institute of Technnology, some NN guy (Steven Bresseler??) https://www.stevens.edu/schaefer-school-engineering-science/departments/computer-science/undergraduate-programs/cybersecurity 48************************************************48 24************************24 # Table of Contents : # $ grep "^#]" "$d_webRawe"'My sports & clubs/SERC systems engng/0_SERC notes.txt' | sed 's/^#\]/ /' ********************* "$d_webRawe"'My sports & clubs/SERC systems engng/0_SERC notes.txt' 06Apr2022 Alexander Kott "Cyber Resilience: a Technical Concept or Vague Desiderata?" 24Feb2021 Donaldson “What Does Digital Transformation Look Like from the C-Suite?” 27Dec2020 Paul Rosenbloom, “Can Graphical Models Provide a Sufficient Basis for General Intelligence?” 24Dec2020 Bill Scherlis SERC TALK: “The Dilemmas of Cybersecurity – Why is Everything Broken?” 24************************24 08********08 #] ??Oct2022 08********08 #] ??Oct2022 08********08 #] ??Oct2022 08********08 #] ??Oct2022 08********08 #] ??Oct2022 08********08 #] ??Oct2022 08********08 #] 09Nov2022 SERC TALKS: “How Can States Develop a Trained Technical Workforce?” ABSTRACT: While the U.S. higher education establishment continues to maintain relatively stable enrollments in spite of a wide range of challenges, the nation’s production of adequately trained technical personnel for a growing number of key jobs in industry and government continues to be inadequate. To fill this gap, states have sought to collaborate with a wide range of industries to create training programs of various kinds and duration to prepare high school graduates for careers in technical occupations sorely needed by many industries. These programs vary in intensity and organization and their funding systems reflect the widely differing conditions of their state sponsors. This talk offers a careful review of a successful program, informed by activities in other states, and highlights the key elements required for success. SPEAKERS: John V. Lombardi, Ph.D is Professor Emeritus at University of Massachusetts Amherst and President Emeritus of University of Florida. Dr. Lombardi is also the former Director of the Center for Measuring University Performance at UMass Amherst-University of Florida serving 2013-2021 (Annual Publication of TARU The Top American Research Universities, 2000-2020). Dr. Lombardi received his MA and Ph.D from Columbia University. Michael Gargano Jr.'s 39-year career in higher education has focused on improving institutional performance indicators common to the field. He is the former President/CEO at St. Vincent's College in Bridgeport, CT. Mr. Gargano has served roles in public and private higher education, including serving as: Provost and Senior Vice President for Academic Affairs for the Connecticut State College and University System; Vice President for Academic, Faculty, and Student Affairs and Dean ad interim at the School of Health Professions at the University of Texas Health Science Center San Antonio; Vice President for Student and Academic Support Services at the Louisiana State University System; Vice Chancellor for Student Affairs and Campus Life at the University of Massachusetts Amherst; and Associate Vice President for Student and Academic Support Services at George Washington University. Mr. Gargano currently serves as a consultant to higher education institutions and Boards of Trustees. This is the second SERC Talk of the Fall 2022 series on “Innovating for STEM Readiness”. These talks are curated and moderated by Dr. William Rouse, SERC Research Council Member, Senior Fellow in the Office of the Senior Vice President for Research at the McCourt School of Public Policy at Georgetown University. +-----+ Q&A All questions (3)My questions (2) Most Recent BH Bill Howell (You) 11:31 AM Lifelong learning - Do these targeted training programs provide a [stronger, better] basis than the normal university programs? Comment BH Bill Howell (You) 11:31 AM How is registration controlled? Do these specialized programs have problems with big surpluses of graduates who are ill-adapted for other jobs? For now you state a very close match of students to available jobs. "... "On average nearly 80% of graduates are employed by the 1st to 3rd quarter upon graduation. ..." Still, maybe its no worse been unemployed with a specialised tarteget degree than with a more general degree? Comment FR Fred Robinson 11:33 AM A great deal of the focus on STEM (local and in general) is concentrated on learning certain disciplines and topics to develop "effective" practitioner skills, but lack an emphasis on development of critical thinking and/or systemic/systems thinking aptitudes, critical to solving "wicked" problems (as in RANGE by David Epstien) vs. "kind" problems (as in PEAK by Anders Eriksson). Where is there room to add the development of the thinking skills habits (e.g., from Waters Foundation) and "divergent thinking" (Ken Robinson) without subtracting from other knowledge development? (ref. also my INCOSE Insight 25:3 article at DOI: 10.1002/inst.12397 ) FR Fred Robinson 11:33 AM A great deal of the focus on STEM (local and in general) is concentrated on learning certain disciplines and topics to develop "effective" practitioner skills, but lack an emphasis on development of critical thinking and/or systemic/systems thinking aptitudes, critical to solving "wicked" problems (as in RANGE by David Epstien) vs. "kind" problems (as in PEAK by Anders Eriksson). Where is there room to add the development of the thinking skills habits (e.g., from Waters Foundation) and "divergent thinking" (Ken Robinson) without subtracting from other knowledge development? (ref. also my INCOSE Insight 25:3 article at DOI: 10.1002/inst.12397 ) This question has been answered live Comment CW Courtney Wright 11:45 AM Stakeholders don't always know what they really want. Have you run into problems with this, where your trained personnel pass verification but fail validation? This question has been answered live 1 Comment CW Courtney Wright 11:52 AM What activities would you like to see from systems engineers to support programs like this? How can we help? 08********08 #] 12Oct2022 SERC TALKS: “Not All Skills Are Created Equal: Agility and Innovation in STEM Education” Speaker: Dr. Richard DeMillo, Professor and Charlotte B. and Roger C. Warren Chair in Computing, School of Cybersecurity and Privacy, Georgia Institute of Technology Abstract: The post-pandemic higher education landscape is very different from the one I wrote about in my 2017 book 'Revolution in Higher Education'. Battles over online education have faded, and new challenges have arisen. This webinar is an update on what has changed and why. Examples will be drawn from the recent Presidential announcement of debt forgiveness for student loans, the continued rise in college tuition, whether a college degree is still relevant, and how the sudden appearance of high-demand interdisciplinary skills change the very idea of academic credentials. It is fair to say that the revolution continues. Moderator: Dr. William Rouse, SERC Research Council Member, Senior Fellow in the Office of the Senior Vice President for Research at the McCourt School of Public Policy at Georgetown University Oct 12, 2022 01:00 PM in Eastern Time (US and Canada) none of fastest growing skills map to an academic domain, eg [pysch ops, container security, cybersecurity], shifting [thresholds, ?subjects?, curriculum] standard curricular areas "owned" by faculty who have taught same thing for 40 years we had addressed same challenge in dotcom bust - students weren't coming in as change too fast ~2001 DeMillo - MIT announced "open courseware", PhysCon - 10 minute videos student didn't want EE degree - they wanted to jump in and DO something doing things in a "stove-pipe" educational system makes less and less sense Question - what about humanities? DeMillo - difficult discussion. Humanities not for a career, discussions Harvey Mudd's ?Valia? Cardo did it well - [English, Math] co-teaching Q K-12 pipeline? is talent coming into Universities? a lot of program VERY bad doesn't like ?15k? independent school boards (probably socialist to force ideas on others) DeMillo - downside of regulation, valuable non-cognitive skills has big impact how do you fit this into a curriculum audio recording /home/bill/SG6/PROJECTS/9_My sports & clubs/SERC - secure content/221012 SERC DeMillo - Agility and Innovation in STEM Education.mp3 08********08 #] 15Jun2022 SERC Secure Cyber Resilient Engineering for the Era of Competition 11:00-12:00 SERC Secure Cyber Resilient Engineering for the Era of Competition (13:00 - 14:00 EDT) /media/bill/Dell2/Website - raw/My sports & clubs/SERC systems engng/SERC Reed 15Jun2022 Secure Cyber Resilient Engineering for the Era of Competition.pdf Ms. Melinda K. Reed Director, Systems Security, STTP @Office of the Under Secretary of Defense for Research and Engineering Ms. Melinda K. Reed is the Director for Systems Security in the Science and Technology Program Protection (STPP) Office in the Office of the Under Secretary of Defense for Research and Engineering (OUSD(R&E)). She brings a wealth of engineering and technical experience to her role, serving as the principal OUSD(R&E) executive for policy, guidance, education, and methods to ensure defense systems perform free of known vulnerabilities and exploitation. Ms. Reed is a recipient of the Under Secretary of Defense (Acquisition, Technology & Logistics) Award for Excellence and the Assistant Secretary of the Navy (Research, Development and Acquisition) Acquisition Excellence Award. &&&&&&&& Howell questions 08********08 #] 15Jun2022 INCOSE AI Explorer - Trustworthiness in AI Systems 08:30-09:45 MDST INCOSE AI Explorer (10:30 – 11:45 EDST) join at http://profile.incose.org regular meetings join INCOSE, then join AI incose.org/ai Tom McDermott : Trust and Explainable AI? In ~2016 Hava Siegelmann UMass Amhearst headed a DARPA program that focused on Explainabe AI. A recent paper on image analysis of malignant cancer woke me up to progress in the area. (Evolutionary computation was key tool for making a tool agnostic solution.) Medical doctors could potentially interact manually with the system, modifying image segments in an intuitive way to understand the results, agree or not, and learn. &&&&&&&& Howell questions Machine Consciousness - Since ~2007, I am happy to see a growing surge of interest in machine consciousness, not as a cute topic, but as a critical component for making advances in Computational Intelligence. So far, I am only comfortable with one concept in the area - that of the late John Tylor of Imperial College in London. Bt there are probably many others, as this area isn't my focus. I haven't thought about machine consciousness from a systems-level, but it's an interesting concept - a like Carl Jungs "social consciousness"? Not sent : @ricardo, @mcdermott -explainability has been a challenge in medical systems forever, and still is today. Black Boxes in highly complex, difficult-to-define, risky situations have long suffered from this. A few overall statistics is inadequate, and like Neural Nets etc, predictability may not be obvious. @barclay - Deep NNs are one of the current targets of projects to provide explainability. They are very difficult to explain. I tend to be a very low-level thinker - the higher levels make no sense to me if it is just rule-based arm-waving. Traditional AI (expert systems, case-bareasoning) are far easier than modern Computational Intellige. But even then, Bernie Widrow's truck backer-upper systems (not expert system - I think it was genetic algo?) were 40 pages of trigonometry etc. That is complete, and simple, but can it really be much use as an explanation? Aniruddha Ranade , shannon.ellsworth@raytheon.com , karla.quintero@irt-systemx.fr , Brian.Junger@vtxco.com , adail.retamal@gmail.com , Luke_rich123@hotmail.com , kenley@purdue.edu , "Rohn, Dennis W. (GRC-LSC0)" , jimc237@verizon.net , nathan.dennis@vtxco.com , nithin.naik@capgemini.com , Thomas Kreitzberg , amanda.muller@ngc.com , dbarnes@spa.com , meek_curran@bah.com , edward.danis@l3harris.com , tkoc3@ford.com.tr , Konstantin Gerasimov , marcosvcardoso@gmail.com , amarkina@gmail.com , Daniel Halloran , Reza Rahdar , Avigdor Zonnenshain , david.ctr.gabello@faa.gov , baoqhs@gmail.com , david.russo@greenpages.com , flavio.oquendo@irisa.fr , lewis.malaver@analog.com , bervin@bechtel.com , sabitoff@gmail.com , mycledinojoseph@johndeere.com , r.wilmot.dunbar@gmail.com , timoszczuk@uol.com.br , Fadl Isa , ornitholographer@googlemail.com , joshua.davidson2@ngc.com , taconna.s.mhoon@medtronic.com , matthew.taylor@mantech.com , esmorri@sandia.gov , jace allen (dSPACE) , ted.kostopoulos@cytiva.com , richard.doornbos@tno.nl , Alper Zihnioglu , steven.bouchired@gmail.com , melvin.e.king8.civ@us.navy.mil , aslaneremre@gmail.com , alessandro.migliaccio@airbus.com , lauradhatt@gmail.com , Bobriant@colostate.edu , john.sliger@gmail.com , demcgowan@gwu.edu , dove@parshift.com , mail@ottmar-bender.de , markgottlieb9@gmail.com , douglas.orellana@mantech.com , dwight.jones@baesystems.com , pjmcgoey@comcast.net , Daniel78lee@gmail.com , Hakan Hökerek , gthibodeau@ieee.org , ajj.jakhel@outlook.com , Joshua Sparber , Mas Muhammad Sukri Bin Masika , Heidi.Davidz@ManTech.com , evdano@comcast.net , remi.riviere@protonmail.com , Ashutosh Kumar , bethany.graham@uk.thalesgroup.com , teka@ue.katowice.pl , munish_kalia-op@yahoo.com , cheney.zhang@siemens.com , eduardo.lopes@incosebrasil.org.br , dejan.stojkovski@nasa.gov , mniitsuma@keio.jp , jrs4g@hotmail.com , rudolph.oosthuizen@up.ac.za , Alan.thomas.bates@gmail.com , mark.m.bennison@mbda-systems.com , abotha@alum.mit.edu , Ricardo Reis , parizi@embraer.com.br , lucas.assis@embraer.com.br , matthew.rice@datasynctech.com , brown.angel.m@gmail.com , Satyanarayana (Satya) Kokkula 08********08 #] 06Apr2022 Alexander Kott "Cyber Resilience: a Technical Concept or Vague Desiderata?" Chief Scientist @US Army Research Laboratory Dr. Alexander Kott serves as the U.S. Combat Capabilities Development Command Army Research Laboratory’s Chief Scientist, reporting to the Director of ARL. Dr. Kott is also the Army Senior Research Scientist (ST) for Cyber Resilience. Prior to becoming the Chief Scientist of ARL, Dr. Kott was the Chief of the Network Science Division at ARL. Earlier, Dr. Kott served as a Program Manager at Defense Advanced Research Projects Agency (DARPA). Kott’s earlier positions included Director of Research and Development at Carnegie Group, Pittsburgh, PA. Dr. Kott earned his PhD in Mechanical Engineering from the University of Pittsburgh, Pittsburgh, PA, in 1989, where he researched AI approaches to invention of complex systems. He received the Secretary of Defense Exceptional Public Service Award. He published over 100 technical papers and served as the co-author and editor of twelve books. Key Kott references Cyber resilience versus Cyber security - important distinction resilience - capacity to recovery quickly from difficulties cyber survivability - closely related to resilience but different Focus - tactical mobile assets important to Army : autonomous, manned, mobile network, forward command post disadvantage - dated, COTS-based, modest SWAP close proximity to adversary - ease of access, penetration physical capture lack of on-board cyber defenders jighly contested network, intermittent connectivity,need to avoid emissions, lack of central monitoring & response Challenges : likelihood of compromise is high cannot take compromised system out of fight (eg tank) recovery by on-board personnel is unlikely future unmanned remonitoring is constrained autonomous recovery is crucial Economics (graph) : costs skyrocket for higher level of risk performance? must encourage resilience resilience Achieving resilience Resiliece By Design (RBD) Resilience By Intervention (RBI) two dimensions : igration and authority RBD tight integration Disadvantage : additional capabil not available RBI external [intervention, authority] ... Disadvantage : delayed response Autonomous Intelligent Agents for cyber reslience image of linked [small 4-rotor drones, soldier, ground robots] - friendly & foe malware Issues : human trust in AI ripe for human-managed agents malware growing, more sophisticated agent perceives, acts complex planning responses and ramifications stealthy execution international working group - conference coming up startup companies limited control - access not always Acceptance & management of risk : (great slide) must accept & manage risk uncomfortable balance risk of consequences if autonomus actionT taken Where we are S&T pursuit, growing literature, How good is your cyber resilience? measure quantitatively cannot improve what you cannot measure compliations may hinder resilience rigorous, repeatable, statstically meaningful James Watt indicator diagram was secret, never patented red teams andntitative assessments are important growing number of defenses increase uncertainty - might decrse resilience How to quantify cyber resilience Kott's book - metrics based, model based Resilience matrix physical, info, .... common idea - area under the curve : cumulative loss of [function, capability, productivity, etc] as adverse event occurs can adapt to improve resilience multiple domains, not just CR scenario dependent How to measure? eg army vehicle - [speefuel consumption, etc] - strange criteria of adveersity? all measures heve deficiencies and concerns measurment - consistent, mission type, monotonicity wrt [defenses, attacks] Where are we now? Army project - 2 year, initial table-top, 3 [,un]manned vehicles, academic partner, data collection, mathematical approaches model - exhibits qualitative behaviou, conceptual insights actual experiment - T of cooling system General lessons articles not readily avilable mission scenarios not obvious [threat, attack]s +-----+ Robert MacKay : [model, test]s of cyber resilience must provide great basis for some aspects of cyber security pririties? Kott: haven't done? Walter Freeman :[model, test]s of cyber resilience must provide great basis for some aspects of cyber security pririties? >> not addreessed Marissa Bigelow : Autonomous agents have limitations and can be insensitive to context / goal changes. How would you recommend monitoring / intervening when agents can operate at beyond human scales? Kott : I assume slower, not so fast Kott : refresh system isn't elegant, but may work Moderator : redundant systems take sensor offline? Chris Sargent : Do you have a methodology to quantitatively evaluate resilience as a function of resilience (recovery) technique complexity? Can RBI be better than RBD in some cases because of a measured complexity? Kott : artillary rounds look for external comm signals, versus onboard AI Richard ?? Saliger? : Q: Referring to slide 38 In case you can disclose this info, what kind of attacks (malware in your model) did lead to the reduction in functionality? Kott : send an email, we decide if can respond +-----+ Howell questions : "slide "autonomous agent fighting for resilience" - is machine consciousness discussed in simple sense - awareness of impact of actions? Kott : should have ability to assess risk of causing harm Area Under curve - what about maximum damage versus overall effect? or different criteria depending on situation? Kott : agrees - area of research & consideration [model, test]s of cyber resilience must provide great basis for some aspects of cyber security pririties? 08********08 #] 24Feb2021 Donaldson “What Does Digital Transformation Look Like from the C-Suite?” Donaldson, SERC 24Feb2021 What Does Digital Transformation Look Like from the C-Suite Digital Transformation works best with new startups, that have nothing to constrain them. Must marry strategy and transformation Excessive silos within organisation Go to outside comps who have suceeded slide7 "... There is no alternative to digital transformation. Visionary companies will carve out new strategic options for themselves… … those that don’t adapt, will fail. ..." Jeff Bezos, Amazon slide8 "... 90% of CEOs believe the digital economy will impact their industry, but less than 15% are executing on a digital strategy ..." MIT Sloan and Capgemini Bounded Rationality • Bounded rationality is the idea that an individual’s rationality is limited in decisions by: • the tractability of the decision problem • the cognitive limitations of the mind • the time available to make the decision • the perceived or actual limits of their view or knowledge (boundaries) • some combination of the above • Decision-makers, in this view, act as satisfiers, seeking a satisfactory (local) solution rather than an optimal (systemic) one. Source: Simon, Herbert; Models of Man My questions : Can you describe examples of where systems engineering itself [failed, suceeded] in overcoming the challenge of "whole systems complexity", and "Bounded rationality"? Perhaps this is not just be an issue for the C-Suite, but for the systems experts as well, especially given the huge number of alternative [concepts, technologies, tools, existing systems]? I like your comment that startup organisations are often more successful with "tranformation + strategy". This sounds a bit like the frequent failure of dominant companies to exploit new [products, services] in their own lines of business, such that new companies then dominate, and there is a continual roll-over of dominant companies over the generations. Could you comment more on limited-department roll-outs, and if some organisations found this to work well, and under which conditions? You mentioned that rarity of systems thinkers - is this a search for the 1 in 10 thousand [leaders, technical, management] people? Weird question - so ignore if too off-track. My hobby science focus since ~1988 is [neural networks, evolutionary computation], where I sometimes see the "new AI" (it's not AI, it's mostly CI - Computational Intelligence) as a demonstration of the [limitations, failures] of [rational, logical, scientific] reasoning, leading into the sucesses of the new techniques which are not [bound, limited] by that. Do you sometimes feel that formal logic is sometimes too limiting with [transformations, strategies]? 08********08 #] 27Dec2020 Paul Rosenbloom, “Can Graphical Models Provide a Sufficient Basis for General Intelligence?” https://sercuarc.org/event/serc-talks-can-graphical-models-provide-a-sufficient-basis-for-general-intelligence Wednesday, April 5, 2017 @ 1:00 pm - 2:00 pm EDT also see http://cogarch.ict.usc.edu Speaker: Paul Rosenbloom, USC Institute for Creative Technologies | CONTACT Sigma is an attempt to build such an architecture from the ground up based on graphical models, a highly efficient, theoretically elegant, and broadly applicable technology for computing with complex multivariate expressions. The goal is to leverage this breadth in blending symbolic high-level cognition with quantitative low-level processing; this theoretical elegance in constructing the diversity of requisite intelligent functionality from interactions among a small very general set of primitives; and this efficiency to build systems capable of practical application. Rosenbloom 05Apr2017 Can Graphical Models Provide a Sufficient Basis for General Intelligence # 29Dec2020 I started watching it GML = Graphical Model model (also probablistic graphical models) RL = multi-agent reinforcement learning Sigma = basic concept SLAM = Simultaneous localization and mapping ToM = rule-based, probablistic and social reasoning +-----+ Introduction Preliminary Definitions - Intelligence “… a very general mental capability that, among other things,involves the ability to reason, plan, solve problems, think abstractly,comprehend complex ideas, learn quickly and learn from experience.” (Editorial in Intelligence with 52 signatories) - General intelligence - What is common across cognitive tasks - Artificial Intelligence “… the scientific understanding of the mechanisms underlying thought and intelligent behavior and their embodiment in machines.” (AAAI) - Artificial general intelligence - The ability of a machine to perform any (human) cognitive task Cognitive Architecture - Model of the fixed structure of a/the mind - Memory, reasoning, learning, interaction, ... - Integration across these capabilities - Supports knowledge and skills above the architecture Overall Desiderata for the Sigma (𝚺) Architecture - A new breed of cognitive architecture that is - Grand unified - Cognitive + key non-cognitive (perceptuomotor, affective, attentive, …) - Generically cognitive - Spanning both natural and artificial cognition - Functionally elegant - Broadly capable yet simple and theoretically elegant - “cognitive Newton’s laws” - Sufficiently efficient - Fast enough for anticipated applications - For virtual humans & intelligent agents/robots that can - Think – Broadly, deeply and robustly cognitive - Behave – Interactive with their physical and social worlds - Learn – Adaptive given their interactions and experience +-----+ SOAR (15 years) to SIGMA projects Overall Desiderata for the Sigma (𝚺) Architecture A new breed of cognitive architecture that is - Grand unified - Cognitive + key non-cognitive (perceptuomotor, affective, attentive, …) - Generically cognitive - Spanning both natural and artificial cognition - Functionally elegant - Broadly capable yet simple and theoretically elegant “cognitive Newton’s laws” - Sufficiently efficient - Fast enough for anticipated applications For virtual humans & intelligent agents/robots that can - Think – Broadly, deeply and robustly cognitive - Behave – Interactive with their physical and social worlds - Learn – Adaptive given their interactions and experience Modular versus Functionally Elegant Memory types : Semantic Episodic Procdural Imagery Working memory : all of above Perception Motor Long-term memory : Declarative Procedural Goal: Advancing elegance, depth and breadth of both science and systems Howell : Robert Hecht-Nielson's Confabulation Theory has all of these Approach: Graphical Architecture Hypothesis Key to success is blending what has been learned from over three decades of independent work in cognitive architectures and graphical models Graphical Models : Efficient computation over multivariate functions by leveraging forms of independence to decompose them into products of simpler subfunctions  Bayesian/Markov networks, Markov/conditional random fields, factor graphs  Solve typically via some form of message passing or sampling  State of the art performance across symbols, probabilities and signals from uniform representation and reasoning algorithm  (Loopy) belief propagation, forward-backward algorithm, Kalman filters, Viterbi algorithm, FFT, turbo decoding, arc-consistency, production match, …  Can support mixed and hybrid processing  Several neural network models map directly onto them Summary Product Algorithm Compute variable marginals (sum/integral-product) or mode of entire graph (max-product)  Pass messages on links and process at nodes  Messages are distributions over link variables (starting w/ evidence)  At variable nodes messages are combined via pointwise product  At factor nodes do products, and summarize out unneeded variables: 12 21 32 ... f (x,y,z)=y2+yz+2yx+2xz m(y)= m(x) x ò ´ f 1 (x, y) =(2x+y)(y+z)= f1(x,y) f2(y,z) Overall Progress on Sigma [JAGI 16] Memory  Procedural (rule) [ICCM 10]  Declarative (semantic/episodic) [ICCM 10, CogSci 14]  Constraint [ICCM 10]  Distributed vectors [AGI 14a]  Perceptual [BICA 14a, AGI 15]  Neural network [AGI 16]  Problem solving  Preference based decisions [AGI 11]  Impasse-driven reflection [AGI 13]  Decision-theoretic (POMDP) [BICA 11b]  Theory of Mind [AGI 13, AGI 14b]  Learning [ICCM 13]  Concept (supervised/unsupervised)  Episodic [CogSci 14]  Reinforcement [AGI 12a, AGI 14b]  Action/transition models [AGI 12a]  Models of other agents [AGI 14b]  Perceptual (including maps in SLAM)  Neural network  Efficiency [ICCM 12, BICA 14b] Overall Progress on Sigma [JAGI 16]  Mental imagery [BICA 11a, AGI 12b]  1-3D continuous imagery buffer  Object transformation  Feature & relationship detection  Perception  Object recognition (CRFs) [BICA 11b]  Speech recognition (HMMs) [BICA 14a, BICA 16]  Localization [BICA 11b]  Natural language  Word sense disambiguation [ICCM 13]  Part of speech tagging [ICCM 13]  Sentence identification [WS 15]  Dialogue [WS 15]  Affect [AGI 15]  Appraisal  Attention  Integration  CRF+Localization+POMDP [BICA 11b]  Rules+SLAM+RL+ToM+VH [IVA 15, WS 15]  SLAM+Appraisal+Attention+VH  SentenceID+Dialogue [WS 15, ICAVCD 16 +-----+ Points taken during video : Functionally elegant versus modular Goal - elegance, depth, breadth of science & systems Cognitive architectures + graphical models [Bayesian, Markov] networks, [Markov, conditional] random fields, factor graphs typically some form of message passing [symbols, probabilities, signals] - (loopy) belief propagation, forward-backward algo, Kalman filters, Viterbi algo, T, turbo decoding, arc-consistency, production match, etc factor graphs - are generalisation of Bayesian graphs Cognitive arch - predicates, conditionals, (Soar-like) nested tri-level control graphical arch - models, piecewise linear functions, gradient-descent learning Soar nested tri-level control - one thing Soar got right [parallel, reactive] layer [serial, iterative] deliberative level [recursive, reflective] layer [bi, tri] models in psycholand science One-state-at-a-time backpropagation Neural networks Appraisal based exploration (curiosity when unfamiliar) INOTS = Immersive naval officer training system, conversational virtual human mind missing new structures and predicates Conclusions - bigger picture of National Human-AI R&D strategic plan timeline of thes2 - slide /42 - interesting : +-----+ The National Artificial Intelligence Research and Development Strategic Plan (Basic R&D segment of Fig. 4) Long-term investments Data aalysiis perception theoretical limitations General AI Scalable AI Human-like AI Robotics Hardware Human-AI collaboration Human-aware AI Human augmentation natural language processing Inferences and visualisations +-----+ JASON (1/17): Perspectives on Research in Artificial Intelligence and Artificial General Intelligence Relevant to DoD 3. The Deep Learning Revolution 5. Areas of Rapid Progress other than Deep Learning 5.1. Reinforcement Learning 5.2. Graphical Models 5.3. Generative Models and Probabilistic Programming Languages 5.4. Hybrid Architectures Sigma is an approach to combining all five! But can AGI based on these five concepts advance more rapidly? Selected Recommendations from report: DoD should both track (via a knowledgeable cadre) and invest in(via a 6.1 research portfolio) the most dynamic and rapidlyadvancing areas of AI, including, but by no means limited to DL.DoD’s portfolio in AGI should be modest and recognize that it isnot currently a rapidly advancing area of AI. … 08********08 #] 24Dec2020 Bill Scherlis SERC TALK: “The Dilemmas of Cybersecurity – Why is Everything Broken?” https://sercuarc.org/event/serc-talks-the-dilemmas-of-cybersecurity-why-is-everything-broken/ Wednesday, November 1, 2017 @ 3:00 pm - 4:00 pm EDT Speaker: Dr. William L. Scherlis, Carnegie Mellon University | CONTACT >> Great talk, over my head of course. Driven by scale-up & complexity auto-immune systems - Linux kernel automatically [mined, repaired] for defects # enddoc