Category Archives: Conferences

Phil 3.12.19

7:00 – 4:00 ASRC PhD



Phil 3.10.19

Learning to Speak and Act in a Fantasy Text Adventure Game

  • We introduce a large scale crowdsourced text adventure game as a research platform for studying grounded dialogue. In it, agents can perceive, emote, and act whilst conducting dialogue with other agents. Models and humans can both act as characters within the game. We describe the results of training state-of-the-art generative and retrieval models in this setting. We show that in addition to using past dialogue, these models are able to effectively use the state of the underlying world to condition their predictions. In particular, we show that grounding on the details of the local environment, including location descriptions, and the objects (and their affordances) and characters (and their previous actions) present within it allows better predictions of agent behavior and dialogue. We analyze the ingredients necessary for successful grounding in this setting, and how each of these factors relate to agents that can talk and act successfully.

New run in the dungeon. Exciting!

Finished my pass through Antonio’s paper

Zoe Keating (May 1) or Imogen Heap (May 3)?

Phil 3.9.19

Understanding China’s AI Strategy

  • In my interactions with Chinese government officials, they demonstrated remarkably keen understanding of the issues surrounding AI and international security. It is clear that China’s government views AI as a high strategic priority and is devoting the required resources to cultivate AI expertise and strategic thinking among its national security community. This includes knowledge of U.S. AI policy discussions. I believe it is vital that the U.S. policymaking community similarly prioritize cultivating expertise and understanding of AI developments in China.

Russian Trolls Shift Strategy to Disrupt U.S. Election in 2020

  • Russian internet trolls appear to be shifting strategy in their efforts to disrupt the 2020 U.S. elections, promoting politically divisive messages through phony social media accounts instead of creating propaganda themselves, cybersecurity experts say.

Backup phone

Work on SASO paper – started

Rachel’s dungeon run is tomorrow! Maybe cross 10,000 posts?

Look at using BERT and the full Word2Vec model for analyzing posts

The Promise of Hierarchical Reinforcement Learning

  • To really understand the need for a hierarchical structure in the learning algorithm and in order to make the bridge between RL and HRL, we need to remember what we are trying to solve: MDPs. HRL methods learn a policy made up of multiple layers, each of which is responsible for control at a different level of temporal abstraction. Indeed, the key innovation of the HRL is to extend the set of available actions so that the agent can now choose to perform not only elementary actions, but also macro-actions, i.e. sequences of lower-level actions. Hence, with actions that are extended over time, we must take into account the time elapsed between decision-making moments. Luckily, MDP planning and learning algorithms can easily be extended to accommodate HRL.

Phil 3.7.19

Day 2 of the TF Dev summit. Worth the money, though much less research-y and more implementation and production-y

Google Cloud has Fedramp certification, which it does see details here.

Live Transcribe

Coral: On Device Transfer learning (paper)

TF 2.0 API \changes and Behavior changes

  • Best practices (link: )
  • Declare variables at the beginning of the code
  • Keras Functional API
    • The Keras functional API is the way to go for defining complex models, such as multi-output models, directed acyclic graphs, or models with shared layers.
  • Autograd can automatically differentiate native Python and Numpy code. It can handle a large subset of Python’s features, including loops, ifs, recursion and closures, and it can even take derivatives of derivatives of derivatives. It supports reverse-mode differentiation (a.k.a. backpropagation), which means it can efficiently take gradients of scalar-valued functions with respect to array-valued arguments, as well as forward-mode differentiation, and the two can be composed arbitrarily. The main intended application of Autograd is gradient-based optimization. For more information, check out the tutorial and the examples directory.
  • JAX is Autograd and XLA, brought together for high-performance machine learning research. With its updated version of Autograd, JAX can automatically differentiate native Python and NumPy functions. It can differentiate through loops, branches, recursion, and closures, and it can take derivatives of derivatives of derivatives. It supports reverse-mode differentiation (a.k.a. backpropagation) via grad as well as forward-mode differentiation, and the two can be composed arbitrarily to any order.
  • Effective TF 2.0: There are multiple changes in TensorFlow 2.0 to make TensorFlow users more productive. TensorFlow 2.0 removes redundant APIs, makes APIs more consistent (Unified RNNsUnified Optimizers), and better integrates with the Python runtime with Eager execution.

Phil 3.6.19

5:00 – ASRC TL

  • Got a lot done on the BAA on the flight yesterday
  • Wrote up a description of LMN and CM for Eric V.
  • Reading more of the Handbook of Latent Semantic Analysis. It’s giving me some good ideas for calculating similarities of posts using Word2Vec and comparing the average vector for each post
  • Antonio got an extension to the 12th. Need to see what he’s up to. Wow, there’s a lot there now. Made some comments about what I’d like to see. I’ll pull down the document to read later
  • Continued to tweak the slides
  • TF Dev conference main sessions today. Breakouts tomorrow.

Phil 1.25.19

7:00 – 5:30 ASRC NASA/PhD

    • Practical Deep Learning for Coders, v3
    • Continuing Clockwork Muse (reviews on Amazon are… amazingly thorough) , which is a slog but an interesting slog. Martindale is talking about how the pattern of increasing arousal potential and primordial/stylistic content is self-similar across scales of the individual work to populations and careers.
    • Had a bunch of thoughts about primordial content and the ending of the current dungeon.
    • Last day of working on NOAA. I think there is a better way to add/subtract months here in stackoverflow
    • Finish review of CHI paper. Mention Myanmar and that most fake news sharing is done by a tiny fraction of the users, so finding the heuristics of those users is a critical question. Done!
    • Setting up Fake news on Twitter during the 2016 U.S. presidential election as the next paper in the queue. The references look extensive (69!) and good.
    • TFW you don’t want any fancy modulo in your math confusing you:
      def add_month(year: int, month: int, offset: int) -> [int, int]:
          # print ("original date = {}/{}, offset = {}".format(month, year, offset))
          new_month = month + offset
          new_year = year
          while new_month < 1:         new_month += 12         new_year -= 1     while new_month > 12:
              new_month -= 12
              new_year += 1
          return new_month, new_year
    • Got a version of the prediction system running on QA. Next week I start something new


Phil 1.24.19

7:00 – 4:30 ASRC NASA/PhD

  • Fake news on Twitter during the 2016 U.S. presidential election
    • The spread of fake news on social media became a public concern in the United States after the 2016 presidential election. We examined exposure to and sharing of fake news by registered voters on Twitter and found that engagement with fake news sources was extremely concentrated. Only 1% of individuals accounted for 80% of fake news source exposures, and 0.1% accounted for nearly 80% of fake news sources shared. Individuals most likely to engage with fake news sources were conservative leaning, older, and highly engaged with political news. A cluster of fake news sources shared overlapping audiences on the extreme right, but for people across the political spectrum, most political news exposure still came from mainstream media outlets.
  • One Simple Trick is now live on IEEE!
  • Antibubbles is going well
  • Work on CHI review. Mention this: Less than you think: Prevalence and predictors of fake news dissemination on Facebook
  • Starting to work on the Slack data ingestion and database population. I really want a file dialog to navigate to the Slack folders. StackOverflow suggests tkinter. And lo, it worked just like that:
    import tkinter as tk
    from tkinter import filedialog
    root = tk.Tk()
    file_path = filedialog.askopenfilename()
  • More beating on the prediction pipeline
    • Load up all the parts of the prediction histories and entries – done
    • Store the raw data in the various prediction tables – done
    • populate PredictedAvailableUDO table – done
    • There’s an error in interpolate that I’m not handling correctly, and I’m too cooked to be able to see it. Tomorrow. interpolatebug

Phil 12.7.18

7:00 – 4:30 ASRC NASA/PhD

Phil 10.15.18

7:00 – ASRC BD

  • Heard about some interesting things this morning on BBC Business Daily – Is the Internet Fit for Purpose?:
    • Future in Review Conference: The leading global conference on the intersection of technology and the economy. New partnerships, projects, and plans you can’t afford to miss. If your success depends on having an accurate view of the future, or you’d like to meet others who are able and motivated to forge action-based alliances, this is the most important conference you will attend. Be one of the thought leaders in the FiRe conversation, analyzing and creating the future of technology, economics, pure science, the environment, genomics, education, and more.
    • Berit Anderson. Created the science fact/fiction magazine Scout, which, interestingly enough, has a discussion space for JuryRoom-style questions
  • More DARPA proposal

Phil 10.4.18

7:00 – 5:30 ASRC MKT

  • Join PCA! Write classified! Done
  • There are 56 work days until Jan 1. My 400 hours is 50 days. So I go full time on research around the 22nd.
  • Got a note from Wayne saying that there were 25 blue sky papers and 3 slots. THat might me expanded to 6 slots
  • Write up notes on “At Home in the Universe” – started
  • Finish speaking notes for BAA – Done
  • Matt found a couple of things that might be good. One is due on October 16th, which is waaaaaaaaaaaaayyyyyyyy too tight.
  • Looked at the Connected Health clearinghouse effort and website. It sounds a lot like a military version of PubMed, with the ability to request reports on demand, plus some standardized reports as well. These reports seem to source back to other agencies like the CDC, with external SMEs.

Phil 10.3.18

7:00 – 5:30 ASRC MKT

  • Finished At Home in the Universe. Really good. I’ll work on writing up notes this evening. The Kindle clippings feature is awesome
  • The stampeding robots paper is up on ArXiv: Disrupting the Coming Robot Stampedes: Designing Resilient Information Ecologies
  • Dopamine modulates novelty seeking behavior during decision making.
  • Need to finish Antonio’s paper, but my sense at this point is to add our work as a discussion of edge conditions that come up in the discussion section?
    • Done. Sent a letter discussing NIST RCS
  • Need to write up the fitness landscape thoughts. One axis is distance to model which is has a decay radius from each agent. Another axis is the price of an item(with future discounting?). Another axis is cost by agent to acquire the item. Cluster behavior emerges from local agents trying to find the best model and acquire the most value? There is also some kind of explicit connection between individuals that needs to be handled (a tanker and a plane have a client-server relationship that requires them to move in a coordinated way)
    • There is also information that is within the agents, and information that is in the environment. There may be other types of information as well.
  • Get Matt rolling on the whitepaper? – done!
  • Watson backend to A2P?
  • Kibitzed Aaron on how to access style sheets
  • Got about halfway through speaking notes on Army BAA

Phil 9.25.18

7:00 – 5:00 ASRC MKT

  • Wayne’s notes from yesterday:
    • Part of the wrapper for this will be why these issues might matter for the iSchool’s research future. I can help with the framing there.
      Yikes, 4 pages in this format? That is nothing!
      Will really have to shave this down to the absolute minimum.
      To that end I think the scenarios get fleshed out in their fullest now to capture all of the ideas and then hacked brutally into 1-2 paragraphs.
      The abstract probably goes to 4 sentences.
      Images stay, but no larger.
      We’ll work this out, but, man, that is barely 1500 words. Who was thinking when they put this together? 😉
  • Want to redo the designed system chart so that the complexity zone is concave – done.
  • More writing. Figured out that cars would be crashing at a rate of 3-4/sec based on 2016 data. Yikes!
  • Worked with Aaron on response to Antonio’s proposal. IEEE Software is a “production” magazine. And a nice marker for production is what kind of libraries are available, because then articles can be written on how to use them.
  • Kate Starbird this Friday! 10:00am – 12:00pm 2119 Hornbake Library South Wing
  • There is a world nomad games

Phil 7.9.18

SASO 2018

Trust in Organizations: Frontiers of Theory and Research

  • Although its importance is readily apparent, the contours of trust in collective contexts are much less obvious. The decision to trust in collective settings is different from, and in many respects more problematic than, decisions about trust that arise in other social contexts. Because of the size and structural complexity of large organizations, for example, individuals do not have the opportunity to engage in the sort of incremental and repeated exchanges that have been shown to facilitate the development of trust in more intimate settings, such as dyadic relationships

Keynote 4 – Heiko Hamann (chair R.P. Würtz, room: FBK – Sala Stringa)

  • Florarobotica
  • Micro and macro behaviors. He’s talking about ants, but I’m interested in UIs. What interfaces/communication channels cause an intelligent agent to perform constructive/destructive/etc behaviors
  • Swarm Robotics: A Formal Approach
  • How do you make decisions on the macroscopic scale?
  • Fast and sloppy vs. slow and accurate. What about ‘hardware acceleration?
  • Kilobot
  • Voter models
  • Majority rule is faster diffusion than voter model, but less accurate.
  • Mixing models does affect the results. Is this explore/exploit as well?
  • Swarm performance over density is a left-skewed normal distribution? Is this because they occupy physical space?
  • Hmmm. Density is a function of dimension
  • Density adaptation. Is it true that people like density to a point? And that online always “feels” different from actual density? What does popularity ranking vs other ranking do to human behavior?
  • subCULTron
  • generic, scalable and decentralized fault detection for robot swarms
    • Robot swarms are large-scale multirobot systems with decentralized control which means that each robot acts based only on local perception and on local coordination with neighboring robots. The decentralized approach to control confers number of potential benefits. In particular, inherent scalability and robustness are often highlighted as key distinguishing features of robot swarms compared with systems that rely on traditional approaches to multirobot coordination. It has, however, been shown that swarm robotics systems are not always fault tolerant. To realize the robustness potential of robot swarms, it is thus essential to give systems the capacity to actively detect and accommodate faults. In this paper, we present a generic fault-detection system for robot swarms. We show how robots with limited and imperfect sensing capabilities are able to observe and classify the behavior of one another. In order to achieve this, the underlying classifier is an immune system-inspired algorithm that learns to distinguish between normal behavior and abnormal behavior online. Through a series of experiments, we systematically assess the performance of our approach in a detailed simulation environment. In particular, we analyze our system’s capacity to correctly detect robots with faults, false positive rates, performance in a foraging task in which each robot exhibits a composite behavior, and performance under perturbations of the task environment. Results show that our generic fault-detection system is robust, that it is able to detect faults in a timely manner, and that it achieves a low false positive rate. The developed fault-detection system has the potential to enable long-term autonomy for robust multirobot systems, thus increasing the usefulness of robots for a diverse repertoire of upcoming applications in the area of distributed intelligent automation.


Satisfy: Towards a self-learning analyzer for time series forecasting in self-improving systems

  • Hyperparameter tuning!
  • Add a meta-learning component (Forecasting module)DSCN0629
  • AutoWEKA (CASHO) – automatically defined search space <— this works!
  • Clustering on the data to try different algorithms on clusters
  • Need to spend some time looking at random forest. It seems to be high value for lower cost
  • Sante-fe institute time-series contest (set A) (set B)

Adaptive Coordination to Complete Mission Goals
Charles Walter, Sarra Alqahtani, and Rose Gamble

  • What Reasonable Guarantees Can We Make for a SISSY System? Kirstie Bellman
    • Trustworthy and knowable
    • Guarantees

Hierarchical Self-Awareness and Authority for Scalable Self-Integrating Systems
Ada Diaconescu, Barry Porter, Roberto Rodrigues Filho and Evangelos Pournaras

  • Hierarchy: Not authority and control, Not distribution. Do mean a multilevel system of different levels of abstraction with feedback loops between them
  • Levels of abstraction allow availability through loss of irrelevant information
  • Executes at different time scales
  • Abstraction goes up, feedback goes down
  • DSCN0632
  • DSCN0633

Security Issues in Self-improving System Integration – Challenges and Solution Strategies Henner Heck, Bernhard Sick and Sven Tomforde

  • Hardly manageable system structures
  • Capabilities
    • Mutual influence detection
    • Mutual dependency detection
    • Emergence detection
    • Self-reflection
  • Am I approaching the optimum?
  • Am I degraded?
  • Additional attack vectors targeting
    • self reflection
    • trust
    • mutual influence
    • adaptation

Improving Security and Interoperability of Interwoven Systems through Rigorous Selective Encapsulation of Critical Physical Resources
Phyllis Nelson

Outriggers and Training Wheels for Cooperating Systems  – Christopher Landauer

  • Kreitman’s Theorem?
  • Training wheels are constraints on a system that prevent catastrophic failures.
  • DSCN0633
  • Efficiency and robustness are direct competitors

A Concept for Proactive Knowledge Construction in Self-Learning Autonomous Systems Anthony Stein, Sven Tomforde, Ada Diaconescu, Jörg Hähner and Christian Müller-Schloer <- nice paper?

  • Reactive knowledge creation = trial and error
  • Proactive behavior = new knowledge created before it is needed
  • MLOC architectures (Multi-Layer Observer/Controller)
  • Drifting distributions leads to changes in the fitness function
  • Suggested NNMF or actually tensor factorization as a way of filling in the tensor

Aspects of Measuring and Evaluating the Integration Status of a (Sub-)System at Runtime
Christian Gruhl, Sven Tomforde, and Bernhard Sick

  • Entity and system-based evaluation

Coopetitive Soft Gating Ensemble
Jens Schreiber, Maarten Bieshaar, André Gensler, Bernhard Sick and Stephan Deist

  • Ensemble estimators?
  • Combinations of weighting and gating to do optimized selection of the best estimator from a collection of estimators that work in particular situations.
  • DSCN0635

Levels of Networked Self-awareness
Lukas Esterle (chair for next years’ SASO)  and John N.A. Brown


Phil 9.6.18

SASO 2018

IEEE (f*) proceedings list (conference should show up here)

Keynote Gábor Vásárhelyi – CollMot Robotics, Budapest, Hungary (chair A. Montresor, room: FBK – Sala Stringa)

  • Multi-level hierarchy in pigeons (gps), wild horses (drone video)
  • Tight turns with pigeons rearrange order. Gentle turns maintain structure
  • Universal rules of collective motion
  • evolution is the best optimizer that they have found
  • Hierarchy aids emergent optimization
  • Drones are exploding in youth. Why? Fashion? Like AI in the ’70’s. Will it then crash?
  • Salable in size and speed
  • Tolerate noise, delay, and error tolerant
  • Controllable meta-unit
  • Drones need to be
    • Extensable
    • real-time
    • sensors
    • communication
    • simulation for dev and debugging
    • algorithms
  • Enemies
    • delay
    • noise
    • inertia,
    • finite acceleration
    • communication errors
    • environment?
  • Communication delays bring self-exited oscillation (repulsion-attraction temporal imbalance)
  • Evolution is the second-best optimization function for all problems
  • Solving decentralized traffic control. A sort of inverse flocking issue
  • Anisotropic repulsion
  • Selective alignment
  • Publications
  • Smaller and lighter (more maneuverable) flocks are more egalitarian. Heavier, smaller groups have hierarchies because there can be trust.

Rule-based utility-driven???
Sona Ghahremani, Christian Medeiros Adriano and Holger Giese

  • Another hyperparameter approach. This one seems quite good
  • Handles
    • Non-linearities
    • Uncertainty from dynamic archetecture
    • Black box models
  • How to select the right model for a class of utility functions prior to deployment
  • Graph grammar rules
  • mRUBiS
    • mRUBiS as an exemplar for model-based architectural self-adaptation is now available at GitHub. It provides a simulator and architectural runtime model of mRUBiS to develop, evaluate, and compare self-adaptation solutions.

Model-Driven Elasticity Control for Multi-Server Queues Under Traffic Surges in Cloud Environments
Venkat Tadakamalla and Daniel Menasce

A Temporal Model for Interactive Diagnosis of Adaptive Systems
Ludovic Mouline, Amine Benelallam, Francois Fouquet, Johann Bourcier and Olivier Barais

  • Might be a way of evaluating fitness tests?
  • At least keep track of the last N (or whole history?) of chromosome so that the principal components can be analyzed.
  • By comparing the states over time, we could find the states adjacent to problem states?

Towards Generic Adaptive Monitoring
Thomas Brand and Holger Giese

  • Goal adaptation (fitness test adaptation?)

Implementing Feedback for ABM Meta-Modeling
Karan Budhraja and Tim Oates

PINCH: Self-Organized Context Neighborhoods for Smart Environments
Chenguang Liu, Christine Julien and Amy L. Murphy

  • Neighborhood mesh networks
  • everything is stored in a small beacon that is cooperatively shared in the neighborhood

A QoS-aware Adaptive Mobility Handling Approach for LoRa-based IoT Systems
Robbe Berrevoets and Danny Weyns

Reins-MAC: Firefly Inspired Communication Scheduling for Reliable Low-Power Wireless
Matteo Ceriotti and Amy L. Murphy

Phil 9.5.18

Marco Barbina – Leonardo SpA – Italy (chair E. Di Nitto, room: FBK – Sala Stringa)

  • Sensors on UAVs
    • Electro-optic
    • Radar-based
    • Hyper-spectral(?)
  • Sensors are not the problem, bandwidth is
  • Selecting the relevant information
  • Attention – which parts of the signal are important
    • Autoencoder NN – it’s like dropout?
    • Split domain stacked cnn. The front half (encoder) NN encodes the data and transmits the “compressed” data from the hidden layer to the back half of the NN.

Goal-aware Team Affiliation in Collectives of Autonomous Robots
Lukas Esterle

  • Teamwork in nature
  • Online multi-task k-assignment
  • stampedes!
  • Observer/follower == explore/exploit
  • They did use the multi-armed bandit framework as a learning system.
  • Real world implementation is intruder detection and response

Optimizing Transitions Between Abstract ABM Demonstrations (video presentation)
Brian Seipp, Karan Budhraja and Tim Oates

  • What is the optimal path that connects two behaviors
  • Traverse a behavior coordinate space
  • Getting from a start image to a goal image. The image can “mean” many things. Agents provide the mapping?
  • Intermediate states are shown
  • Noise added to guarantee convergence?

Self-organized Resource Allocation for Reconfigurable Robot Ensembles
Julian Hanke, Oliver Kosak, Alexander Schiendorfer and Wolfgang Reif

  • Search – determine dangers
  • Continual observe
  • React
  • This is a fitness space analysis based on optimal configuration of robots for a given task. It’s kind of a distributed traveling salesman problem combined with a market and a bidder as a way of merging the distributed solution. I wonder if it is computationally better than a GA solution. It could be a way of looking at hyperparameter tuning?
  • Not clear that it can handle a floating point value, like fuel level
  • The goal is for the allocation to be fast

Panel: The Future of Autonomic Computing and Self-* Systems:

  • Kristie Bellman – some success but in isolated, hard to define areas. Build foundation of principals that can be used to frame development
  • Ada Diaconescu – Efficiency-oriented or creativity-oriented self* systems
  • Simon Dobson – We have turned a lot of fields into computation. But – AI told us nothing about how people play Go. We’re trying to build, not describe
  • Xiaohui (Helen) Gu – AIOps, bridging the gap between academia and industry
  • Arif Merchant – (Google) Missing holistic, end-to-end solutions. First and follow-on principals.
  • Danny Weyns – Be less ad-hoc and more scientific. Address threats to validity. Solid foundations.
  • I talked a bit about the continuum from the low-hanging fruit of autonomic systems to the difficult task of creating self-aware systems.

A Macro-Level Order Metric for Self-Organizing Adaptive Systems
David King and Gilbert Peterson

  • critical and stable states, based on local and global entropy
  • I think that he is talking about “personalized” entropy rather than “local” entropy.

A Self-Organized Learning Model for Anomalies Detection: Application to Elderly People
Nicolas Verstaevel, Carole Bernon, Jean-Pierre Georgé and Marie-Pierre Gleizes

  • Real time detection of abnormal behaviors by using feedback from the medical staff.
  • Detect anomalies through a linear regression of disparity values
  • Is this another hyperparameter system? It could be looking for sensitivity of the system to hyperparameter

Risk-based Testing of Self-Adaptive Systems using Run-Time Predictions
André Reichstaller and Alexander Knapp

  • Uses machine learning to explore the space that includes errors. It is a way of finding major(?) features in the state space

Play with Generated Adversarial Networks (GANs) in your browser!

  • First, we’re not visualizing anything as complex as generating realistic images. Instead, we’re showing a GAN that learns a distribution of points in just two dimensions. There’s no real application of something this simple, but it’s much easier to show the system’s mechanics. For one thing, probability distributions in plain old 2D (x,y) space are much easier to visualize than distributions in the space of high-resolution images.