Variational Autoencoders For Image GenerationJason Anderson
Meetup: https://www.meetup.com/Cognitive-Computing-Enthusiasts/events/260580395/
Video: https://www.youtube.com/watch?v=fnULFOyNZn8
Blog: http://www.compthree.com/blog/autoencoder/
Code: https://github.com/compthree/variational-autoencoder
An autoencoder is a machine learning algorithm that represents unlabeled high-dimensional data as points in a low-dimensional space. A variational autoencoder (VAE) is an autoencoder that represents unlabeled high-dimensional data as low-dimensional probability distributions. In addition to data compression, the randomness of the VAE algorithm gives it a second powerful feature: the ability to generate new data similar to its training data. For example, a VAE trained on images of faces can generate a compelling image of a new "fake" face. It can also map new features onto input data, such as glasses or a mustache onto the image of a face that initially lacks these features. In this talk, we will survey VAE model designs that use deep learning, and we will implement a basic VAE in TensorFlow. We will also demonstrate the encoding and generative capabilities of VAEs and discuss their industry applications.
Variational Autoencoders For Image GenerationJason Anderson
Meetup: https://www.meetup.com/Cognitive-Computing-Enthusiasts/events/260580395/
Video: https://www.youtube.com/watch?v=fnULFOyNZn8
Blog: http://www.compthree.com/blog/autoencoder/
Code: https://github.com/compthree/variational-autoencoder
An autoencoder is a machine learning algorithm that represents unlabeled high-dimensional data as points in a low-dimensional space. A variational autoencoder (VAE) is an autoencoder that represents unlabeled high-dimensional data as low-dimensional probability distributions. In addition to data compression, the randomness of the VAE algorithm gives it a second powerful feature: the ability to generate new data similar to its training data. For example, a VAE trained on images of faces can generate a compelling image of a new "fake" face. It can also map new features onto input data, such as glasses or a mustache onto the image of a face that initially lacks these features. In this talk, we will survey VAE model designs that use deep learning, and we will implement a basic VAE in TensorFlow. We will also demonstrate the encoding and generative capabilities of VAEs and discuss their industry applications.
Presentación elaborada y compartida por George Siemens en su conferencia en Buenos Aires, invitado por Fundación Telefónica de Argentina, el 12 de septiembre de 2012.
This is a talk about activity systems analysis and its application for design research. This talk was prepared for students and faculty at Florida State University.
Places for News: an exploration of context and situated methodsYuval Cohen
Dissertation project completed for MSc Human-Computer Interaction at UCL Interaction Centre (UCLIC). Explores contextual factors that affect news consumption, and technologies that can be used to research them. Supervised in part by the BBC.
This talk introduced staff at University College Borås to an approach for teaching social media literacies that I was piloting with a group at the IT Technics University, Gothenburg, Sweden.
3. Distributed cognition: toward a new foundation for human-computer interaction research JAMES HOLLAN EDWIN HUTCHINS DAVID KIRSH University of California, San Diego
4. Who are the authors? James D. Hollan Edwin Hutchins David Kirsh
5. James D. Hollan Experience Ph.D in cognitive psychology; Postdoc in artificial intelligence; Research faculty; Intelligent Systems Group at UCSD and Future Technologies Group at NPRDC; Chair of the CS in University of New Mexico (1993); Professor of Cognitive Science at UCSD (1997) Research interest across cognitive ethnography, distributed and embodied cognition, multimodal interaction, information visualization, etc.
6. Edwin Hutchins The father of modern Cognitive Ethnography A strong advocate of the use of anthropological methods The student of the cognitive anthropologies Roy D’Andrade Early work involved the relationships among language, culture, and thought As a postdoc, worked for US Navy, used insights from first-hand ethnographic studies to build training systems for radar navigation Cognition in the Wild (1995)- distributed cognition Moved to commercial aviation Former department head of cognitive science at UCSD Currently, with James Hollan, runs the Distributed Cognition and Human Computer Interaction Lab at UCSD
7. David Kirsh Experience BA from University of Toronto in philosophy and economics Ph.D from Oxford University on foundations of cognitive science A research scientist at MIT Artificial intelligence Lab (1984-1985) Professor at Dept. of Cognitive Science at UCSD and lead interactive Cognition Lab Research interests include artificial intelligence, situated cognition, philosophy of mind and science, interactive design, interactive environment, etc. Publications: Representation and rationality: foundations of cognitive science (1983)
8. Information about this paper ACM Transactions on Human-Computer Interaction: Special Issue on Human-Computer Interaction in the New Millenium. Cited by 742
9. Look at it from big picture.. Single computer networked computers a complex world of information Human information processing psychology Distributed cognition (Hutchins, 1995) Cognition distributed across individuals and artifacts in a social-cultural and technical systems A cognitive theory to understand interaction A framework as a new foundation for HCI How this framework works
10.
11. The range of cognitive eventsTraditional cognitive theory Individuals Cognitive events inside individual actors Distributed cognition Looks for cognitive process a broader class of cognitive events (eg. Airline cockpit)
12. Characteristics of Distributed cognition Socially distributed cognition Distributed across the members of a social group Embodied cognition Involve coordination between internal and external structures Culturally embedded cognition Culture shapes the cognitive processes of systems that transcend the boundaries of individuals
13. An integrated research framework Theory: Distributed cognition Methods: ethnographic observation + experiments Products: digital work materials + collaborative workspaces
14. Examples…design of digital work materials Ship navigation Airline cockpit automation Direct manipulation History-enriched digital objects But these research programs did not exist at the time of this paper..
15. Conclusion All cognition can be fruitfully observed as occurring in a distributed manner well- suited to understanding the complex networked world of information and computer-mediated interactions To inform the design of digital work materials an collaborative workplaces
18. Context Mark Ackerman At time of publication, was an an associate professor in Information and Computer Science at the University of California, Irvine. Currently a professor of EE and CS and School of Information at the University of Michigan Christine Halverson Social Computing researcher at IBM Formerly cognitive scientist in the CHI Center at SRI International, Menlo Park CA The Paper Appeared in the Communications of the ACM in Jan. 2000 ACM lists 15 citations
19. Defining Organizational Memory (OM) Despite 10 years of research, OM has different, sometimes conflicting definitions and little supportive empirical research, thus OM as a concept must result from studies within the context of everyday use An empirical study Ethnography involving telephone helpline group at CyberCorp Analyzed with a more cognitive approach Specific example of database lookup
20. Key Points/Findings There is no monolithic repository of OM Memories as a set of artifacts and processes Even simple tasks involve complicated distributed memory A single agent’s process is simultaneously embedded within several organizational processes Individual memories have mixed provenance Memories are transferred and reused in the form of boundary objects They go through a process of decontextualization and recontextualization Proper consideration should be given about how a memory may be reused.
21. Jacob T. Biehl, William T. Baker, Brian P. Bailey, Desney S. Tan, Kori M. Inkpen, and Mary Czerwinski Presenter: Anamary Leal "IMPROMPTU: A New Interaction Framework for Supporting Collaborationin Multiple Display Environments and Its Field Evaluation for Co-locatedSoftware Development“
22. Cited by 27 Downloaded 223 times At CHI Conference 2008 Impact of Paper
23. Jacob T. Biehl Interests: Human-Computer Interaction, Software Engineering, Programming Languages From: University of Illinois, Urbana Currently at: Research Scientist at FX Palo Alto Laboratory Fact: Getting more liberal by the minute William T. Baker Interests: Supercomputing for astronomy, security planning, automated build system, middleware development From: University of Illinois, Urbana Currently at: Research Programmer at NCSA Fact: Create a diagram from ASCII About the Authors
24. Brian P. Bailey Interests: developing interactive tools that foster human creativity; systems that improve interruption management, and user interfaces for multiple display environments From: University of Illinois, Urbana Currently at: Assistant Professor Desney S. Tan Interests: Human-Computer Interaction, Physiological Computing, and Healthcare From: Microsoft Research Currently at: Senior Researcher in the Visualization and Interaction, Microsoft Research Fact: Serving as General Chair for CHI 2011 in Vancouver, BC About the Authors
25. Kori M. Inkpen Interests: CSCW for home, work, education, healthcare and fun From: Microsoft Research Currently at: Microsoft Research, researcher in the Visualization and Interaction Group Fact: Conference Co-Chair, ACM CSCW 2010, Savannah, GA Mary Czerwinski "Interests: Human-Computer Interaction, Visual Attention, Task Switching, User Interface Design, Information Visualization, Groupware, Ubiquitous Computing, Spatial Cognition, Novel Interaction Techniques" From: Microsoft Research Currently at: Research Area Manager of the Visualization and Interaction (VIBE) Research Group Fact: Likes The Daily Show, The Colbert Report About the Authors
26. IMPROMPU: (IMPROving MDE’s Potential to support Tasks that are genUine New interaction framework that handles sharing in team problem solving and discussion, multiple-display environments Lightweight: works with off-the-shelf applications Evaluated in 3 week field study with Microsoft software development teams Main Argument
27. Build a system that supports the benefits of shared, multi-display environments "to understand how groups leverage MDEs to perform their activities and the resulting impact....for real activities in authentic settings." Goals
28. All applications can be: do not show or share show share Collaborator Dock Shared Screen Dock Designing for Group Work
29. 2 teams, 3 weeks of observations, user logs, and user feedback shared a wide variety of apps Ways of usage: person to person, with verbal face to face or phone multitasking group: using the shared display to show image on personal device Field Study Results
30. 2 kinds of uses: activate as needed run always Physical movement of objects to personal devices decreased by 60-40% Most features used, though not frequently Field Study Results
31. Find better data in terms of activities of replicated windows Merge Collaborator into more information Give user sharing controls regardless of device Recommendations
32. IMPROMPTU: "to support opportunistic, short-lived collaborative engagements" Flexible, lightweight applications encourage better collaboration for multi-display environments Take Home Message
34. Distributed cognition in an In Cognition and Communication at Work airline cockpit Cited:406 33
35. First Author Current Projects: 1. How do airlines outside the US use Boeing airplanes? What can be done to design airplanes for the world's pilots? 2. Light Jet Training for Corporate Aviation Edwin HutchinsSpent his entire academic career trying to understand human cognition in social, cultural and material context. Department of Cognitive ScienceUniversity of CaliforniaSan Diego 34
36. Department of Communication, Aalborg University, DENMARK Second Author Tove Klausen 35
37. Study Background Information processing in the distributed system is: a propagation of representational state across representational media. Keyword: Information, State, Media 36
38. Simulation Hypothesis Cognitive labor is socially distributed Flying a modern jet transport can not be done by an individual acting alone 37
50. Simulation Tasks Planning to climb higher altitude PLAN Redundant read backs for error checking CHECK Intersubjectivity as a basis for communication COMMUNICATE Set radio frequency ACT Distribution of labor again INTER-ACT Computation by propagation and transformation of representational state COMPUATE Intersubjectivity and distribution of storage again COMMUNICATE Firewall thrust ACT 41
51. Simulation Results Information flow is bi-direction Information flow is transparent to all crew members, instruments, and air traffic control There is often a misunderstanding among crew members Individual cognition is the fundamental of the distributed cognition 42
52. Simulation Implication? Airline safety is a system engineering problem Have to consider the overall system to minimize the risk of accident Distributed cognition -> increase redundancy 43
53. Distributed cognition: hardcore Science? Challenges: Distributed ~= Complex ~= Hard to deal with Solutions: Can borrow system engineering methods: data flow diagram, decision matrix, Pugh method 44
54. Sullenberger, a possible opposite example of distributed cognition? Media: this man have saved 155 lives, 1.15.2009 45
55. We almost forget this man Jeffrey Skiles, First Officer of flight 1549 46
56. Questions for Discussion: Under what paradigm do the various paper’s approaches to distributed cognition fall? What is “cognitive ethnography,” and is it somehow different or “better” than normal ethnography? How do organizational memory and distributed cognition relate? Are they just different ways of describing the same thing, or are they somehow distinct?
57. Questions for Discussion: Are CSCW and its tools, such as IMPROMPTU, just a subset of distributed cognition? Intersubjectivity results in some dated artifacts, for example, the floppy disk as the universal save icon. Is this phenomenon counter to progress, or is it just a quirk? Does distributed cognition theory put us in a better position to tackle wicked problems? Or does it itself contain wicked problems? Today, people are constantly documenting everything, with photos, status updates, etc. What ramifications might this have for distributed cognition?
Editor's Notes
David Kirsh (1954 (?)) is a Canadian cognitive scientist, and Professor at University of California, San Diego (UCSD), where he heads the Interactive Cognition Lab[1].He received his BA from the University of Toronto in 1976 and his D.Phil. from Oxford University[2] in 1983 with the thesis Representation and rationality : foundations of cognitive science. Prior to arriving at UCSD, he spent five years as a research scientist at the MIT Artificial Intelligence Laboratory from 1984 to 1989. Since 1989 he is Professor at Deptarment of Cognitive Science at University of California, San Diego (UCSD). Since 1989 he is director of the Interactive Cognition Lab as well.His research interests include interactive design, collaborative environments, cognitive aspects of multimedia design, information architecture, attention management and human-computer interaction.I received my BA from the University of Toronto in philosophy and economics, my D.Phil from Oxford University on foundations of cognitive science, and I spent five years at the MIT Artificial Intelligence Lab as a research scientist. I am a Professor of Cognitive Science at the University of California at San Diego. Although my official areas of specialization are artificial intelligence, situated cognition, philosophy of mind and science, and foundations of cognitive science, I have been working for some years now on cognitive engineering and how to better design highly interactive environments.Kirsh published several books and articles. A selection:1983. Representation and rationality : foundations of cognitive science. Thesis D.Phil. University of Oxford1992. Foundations of artificial intelligence. Edited by David Kirsh Cambridge, Mass ; London : MIT Press.2003. Proceedings of the Twenty-Fifth Annual Conference of the Cognitive Science Society : July 30-August 2, 2003, Park Plaza Hotel, Boston, Massachusetts, USA. edited by Richard Alterman and David Kirsh. Cognitive Science Society (U.S.). Conference (25th : 2003 : Boston,
an integrated framework for research in order to design the intellectual workspace, theories that view human-computer interaction within larger sociotechnical contexts and a theory-based framework are needed. this framework combines ethnographic observation and controlled experiments as a basis for theoretically informed design of digital work materials and collaborative workplaces. Elements shown in figure 1: distributed cognition theory identifies a set of core principles that widely applied and use these principles subsequently to identify a class of phenomena that merit ethnographic observation and documentationcognitive ethnography: has methods for observing, documenting, and analyzing some phenomena, like information flow, cognitive properties of systems, social organizations, and cultural processes. "ethnographically natural" experiment, it happens in a real-world setting. it focuses on understanding cognitive processes enacted in the naturally situated activity. it seeks to determine what things mean to people in an activity through real-world observation and document this means. experiments: make the impact of changes (in the naturally occurring parameters) more precise. (through experimental control) an experiment is another socially organized context for cognitive performance. they happen in settings in which people make use of a variety of materials and social resources. except observing in the real-world setting, we can set about designing more constrained experiments which test specific aspects of certain behavior. so the principles, ethnography, and experiment mutually constrain each other and offer information on the design of work materials. work materials are part of workplaces and constitutes important changes in the DC environment. so the introduction of a new work material itself is a form of experiment, which allows us to test and revise the theory. Relations: ethnographic observation suggests experiments, and then the findings of experiments could refine theory of DC and then improve design, and design process creates new tools for workplaces, there are new structures and interactions to study. this loop forms iterative process to successive refine theory, methods, and products.