Monthly Archives: February 2020

Phil 2.28.20

7:00 – ASRC GOES

AirSim is a simulator for drones, cars and more, built on Unreal Engine (we now also have an experimental Unity release). It is open-source, cross platform, and supports hardware-in-loop with popular flight controllers such as PX4 for physically and visually realistic simulations. It is developed as an Unreal plugin that can simply be dropped into any Unreal environment. Similarly, we have an experimental release for a Unity plugin.

  • Added notes for the dissertation revisions
  • Working on the GVSETS paper – meeting at 3:00. Got everything into SVN and coordinated across machines.
  • Got Deep Learning with Tensorflow2 and Keras to start boning up on before the conference
  • Need to set some time aside for dissertation revisions
  • Keyword search for Shakespeare
  • Still need to fix the race conditions on file write and directory change
  • IRAD meeting. Signed up for Sim as a service, and exploit spaces white paper. Got John to pay for an Overleaf account

Phil 2.27.20

ASRC GOES 7:00 – 8:00

GOES

PhD

  • Meeting with Wayne. I got the (preliminary)
    • Combine the Limitations and Future Work chapters into a new chapter that explores where my research has landed. I think that a beachhead analogy might work here.
      • Preliminary, early work that lives in the space between computational sociology and HCC (socio-technical, from From Keyword Search to Exploration). Add discussions about wormholes, weathermaps, and maps that connect distant places, like air travel maps.
      • Diversity science
      • Gathering data on online consensus in various sized groups and within different cultures
      • Extending simulations into human belief spaces, such as with GPT-2 agents
    • Add a “my hut, revisited” section to the contributions that discusses the contributions from the perspective of the spaces defined by Kauffman, Martindale, and Bacharach in particular, but also the literature in general

Tie into this as well

Some general reworking of the contributions text to reflect the slides: This also requires expansion of the current text in CH11. In particular slides 49-52 are a stronger synopsis than is provided in your CH10 discussion. Slides 53-56 are a more thoughtful framing of your contributions, both theoretical and practical (can you weave this back in your literature from CH2 to show specific knowledge contribution?). The “bookend” revisiting ofthe trustworthy anonymous citizen journalism was also more effective in the presentation than the document. Consider capturing this.

Phil 2.26.20

8:00 – 6:00 ASRC GOES

PhD

  • Successfully defended, so I’m a PhD now!
  • Need to handle graduation paperwork by the end of the month

ASRC

  • Add thread code (from here) back into scratches – done
  • Try out the multi-gpu code – it works!
    • But there is a race condition for the writing of the best and eval
    • Trying to create a smaller test case to break things and work out a fix
    • found a nice lorem ipsum package
  • Start writing up two pager for proposal
  • Meeting with John D and T about APG workshop paper. Sent T a bunch of nores later
  • Mission Meeting
    • Went over VPN, and chatted with Vadim, who’s still having joint constraint problems

Phil 2.19.20

7:00 – 8:00 ASRC GOES

disinfoOps
  • Defense practice and tweaking
  • Continue setting up workstation
    • Java – done
    • Python – done
      • Pandas3d
      • Panda
      • Scikit-learn
      • TF 2.0
      • etc
    • TortoiseSVN – done
    • WinSCP – done
    • PuTTY – done
    • XAMPP – done
    • gVIM – done
    • MikTex – done
    • TexStudio – done
    • Adobe Creative Cloud – done
      • Acrobat- done
      • Illustrator- done
      • Photoshop – done
    • Intellij
    • Office- done
    • Project
    • Chrome- done
    • FF
    • GitHub desktop
    • Set up non-admin user
  • Mission Drive meetings

Phil 2.19.20

8:00 – 7:00 ASRC GOES

  • I have a cold. Yuck!
  • How to build a brain from scratch
    • This advanced option course discusses the search for a general theory of learning and inference in biological brains. It draws upon diverse themes in the fields of psychology, neuroscience, machine learning and artificial intelligence research. We begin by posing broad questions. What are brains for, and what does it mean to ask how they “work”? Then, over a series of lectures, we discuss parallel computational approaches in machine learning/AI and psychology/neuroscience, including reinforcement learning, deep learning, and Bayesian methods. We contrast computational and representational approaches to understanding neuroscience data. We ask whether current approaches in machine learning are feasible and scaleable, and which methods – if any – resemble the computations observed in biological brains. We review how high-level cognitive functions – attention, episodic memory, concept formation, reasoning and executive control – are being instantiated in artificial agents, and how their implementation draws upon what we know about the mammalian brain. Finally, we contemplate the outlook for the future, and whether AI will be “solved” in the near future.
  • Start writing GVSETS simulation white paper for March 3
  • Finish setting up workstation
    • Soooooo, that turned out to be a bigger problem. The way that admin is set up is that there are two levels, an “install software” level, and an “install hardware level”. I had been given the former. When I installed the new SSD, it had to be formatted with NTFS. I could launch the device manager, but could not access the drive. I took the box down to HQ, and their IT folks were mystified. I stopped by the shop that had fixed the power supply, and for another $50, they put the drive in another machine, formatted and named it. Then it worked. Wheee.
    • I then start re-installing software and find that although I can log in as admin, I can’t run anything as admin. That requires going back to HQ so that they can set permissions on the local network, and install a VPN. Now our local network is horrible. There is some kind of whitelist/blacklist filter that is so slow that web pages often timeout rather than load.
    • At this point, I decided that it was easier to reset the machine and start over. So I’ll be doing that for a few days. Today, I got the NVIDIA and CUDA drivers installed.
  • Verify TF and multi-gpu optevolver
  • ML Seminar

Phil 2.17.20

Pinged Aaron about getting together today – 3:00

  • Went through the talk he says it seems pretty solid
  • Also dropped by Don’s office to say hi

Generated a new map where a stampede stops when it hits the edge

Played around with the “Curse of Dimensionality” slide

Added some background on fact-checking to the RQ slide

Start fixing Alienware

  • Mount drive – done
  • Plug everything in and connect network – done
  • Copy files
  • Fix environment variables

Online Conspiracy Theories: The WIRED Guide

  • Everything you need to know about George Soros, Pizzagate, and the Berenstain Bears.

 

Phil 2.16.20

Bringing Stories Alive: Generating Interactive Fiction Worlds

  • World building forms the foundation of any task that requires narrative intelligence. In this work, we focus on procedurally generating interactive fiction worlds—text-based worlds that players “see” and “talk to” using natural language. Generating these worlds requires referencing everyday and thematic commonsense priors in addition to being semantically consistent, interesting, and coherent throughout. Using existing story plots as inspiration, we present a method that first extracts a partial knowledge graph encoding basic information regarding world structure such as locations and objects. This knowledge graph is then automatically completed utilizing thematic knowledge and used to guide a neural language generation model that fleshes out the rest of the world. We perform human participant-based evaluations, testing our neural model’s ability to extract and fill-in a knowledge graph and to generate language conditioned on it against rule-based and human-made baselines. Our code is available at this https URL.

The Obligation To Experiment

  • Tech companies should test the effects of their products on our safety and civil liberties. We should also test them ourselves.

Ran through the presentation with David. He pointed out that stampedes bouncing off the edge of the environment look like flocking, so I generated a new map where the stampede gathers and runs off the edge

flock_2_stampede_map_legend

Phil 2.14.20

7:00 – 8:30 ASRC GOES

This document describes the Facebook Full URL shares dataset, resulting from a collaboration between Facebook and Social Science One. It is for Social Science One grantees and describes the dataset’s scope, structure, fields, and privacy-preserving characteristics. This is the second of two planned steps in the release of this “Full URLs dataset”, which we described at socialscience.one/blog/update-social-science-one.

Judging Truth

    • Deceptive claims surround us, embedded in fake news, advertisements, political propaganda, and rumors. How do people know what to believe? Truth judgments reflect inferences drawn from three types of information: base rates, feelings, and consistency with information retrieved from memory. First, people exhibit a bias to accept incoming information, because most claims in our environments are true. Second, people interpret feelings, like ease of processing, as evidence of truth. And third, people can (but do not always) consider whether assertions match facts and source information stored in memory. This three-part framework predicts specific illusions (e.g., truthiness, illusory truth), offers ways to correct stubborn misconceptions, and suggests the importance of converging cues in a post-truth world, where falsehoods travel further and faster than the truth.

       

 

  •  Dissertation
    • Practice! 52 minutes, 57 seconds
    • Maybe meeting with Wayne? Nope
  • Pack, move, unpack, setup
    • Bring ethernet cables! done
    • Moved out – done
    • Moved in – not done, but ready to unpack
  • Recovered my information for GSAW and TFDev
  • Write quick proposals for:
    • cybermap – done
    • Synthetic data as a service – done
    • White paper – kinda?

Phil 2.13.20

7:00 – 4:00 ASRC (charge number?)

  • AI/ML workshop
  • Color code timeline – done
  • Pick up computer? – done.
    • It turns out that the alienware OEM power supply uses standard connectors in a non-standard way. When I had to use the OEM SATA low-profile connector, I tripped the power supply and also blew out the HD. Ordered a replacement SATA SSD
    • Rebuild travel folders
    • Copy laptop’s d: dev and program files folders onto new SSD

Phil 2.12.20

7:00 – 8:00pm ASRC PhD, GOES

  • Create figures that show an agent version of the dungeon
  • Replicate the methods and detailed methods of the cartography slides
  • Text for each group by room can be compared by the rank difference between them and the overall. Put that in a spreadsheet, plot and maybe determine the DTW value?
    • Add the sim version of the dungeon and the rank comparison to the dissertation
  • Put all ethics on one slide – done
  • Swapped out power supply, but now the box won’t start. Dropped off to get repaired
  • Corporate happy hour

Phil 2.11.20

7:00 – 9:00 ASRC GOES

The brains of birds synchronize when they sing duets

  • When a male or female white-browed sparrow-weaver begins its song, its partner joins in at a certain time. They duet with each other by singing in turn and precisely in tune. A team led by researchers from the Max Planck Institute for Ornithology in Seewiesen used mobile transmitters to simultaneously record neural and acoustic signals from pairs of birds singing duets in their natural habitat. They found that the nerve cell activity in the brain of the singing bird changes and synchronizes with its partner when the partner begins to sing. The brains of both animals then essentially function as one, which leads to the perfect duet. (original article: Duets recorded in the wild reveal that interindividually coordinated motor control enables cooperative behavior)

Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering

  • Top-down visual attention mechanisms have been used extensively in image captioning and visual question answering (VQA) to enable deeper image understanding through fine-grained analysis and even multiple steps of reasoning. In this work, we propose a combined bottom-up and top down attention mechanism that enables attention to be calculated at the level of objects and other salient image regions. This is the natural basis for attention to be considered. Within our approach, the bottom-up mechanism (based on Faster R-CNN) proposes image regions, each with an associated feature vector, while the top-down mechanism determines feature weightings. Applying this approach to image captioning, our results on the MSCOCO test server establish a new state-of-the-art for the task, achieving CIDEr / SPICE / BLEU-4 scores of 117.9, 21.5 and 36.9, respectively. Demonstrating the broad applicability of the method, applying the same approach to VQA we obtain first place in the 2017 VQA Challenge

 

  •  Defense
    • Need to think about how to discuss maps like the T-O and belief space maps (flocking and stampeding projections?) are attention maps as well. Emphasizing well-triangulated but less-attended areas is a potential good. Compare to how maps opened up areas for exploration and exploitation, but this is constructive and not extractive
  • Admin -done
  • Walkthrough of Aaron’s slides
    • Showed him how to outline boxes and reduce the filesize
  • Shimei’s group
    • Walkthrough of the slides
    • Strengthen the connection between the sim and the human study

Phil 2.9.20

In Data Voids: Where Missing Data Can Easily Be ExploitedGolebiewski teams up with danah boyd (Microsoft Research; Data & Society) to demonstrate how data voids are exploited by manipulators eager to expose people to problematic content including falsehoods, misinformation, and disinformation.

Data voids are often difficult to detect. Most can be harmless until something happens that causes lots of people to search for the same term, such as a breaking news event, or a reporter using an unfamiliar phrase. In some cases, manipulators work quickly to produce conspiratorial content to fill a void, whereas other data voids, such as those from outdated terms, are filled slowly over time. Data voids are compounded by the fraught pathways of search-adjacent recommendation systems such as auto-play, auto-fill, and trending topics; each of which are vulnerable to manipulation.

Persuading Algorithms With an AI Nudge Fact-Checking Can Reduce the Spread of Unreliable News. It Can Also Do the Opposite.

Tesla Autopilot Duped By ‘Phantom’ Images: Researchers were able to fool popular autopilot systems into perceiving projected images as real – causing the cars to brake or veer into oncoming traffic lanes.

Phil 2.7.20

7:00 – 5:00 ASRC GOES

tacj

  •  Dissertation
    • Fixed some math
  •  Defense
    • Added a math slide for agent calculations – done
    • Need to add the sim justification slide, based on Aaron’s comments last night – done
    • Drop off signed papers at graduate school this morning – done
    • Working on the abstract – done and submitted!