Monthly Archives: April 2022

Phil 4.29.2022

Book

  • Finished Belief is a Place. Currently at 76k words
  • Cornell Press has a list of editors and how to contact them???

SBIRs

  • JSFC Meetings. First one seemed to go well. Second one was less focused
  • Would this work for entropy measures? Richardson–Lucy deconvolution
  • More code execution. Should be able to get init, step, and terminate working today and print the live DataDictionary – DONE!
  • Need to connect to the SimAccel repo – Done! Changed to scr directory to simaccel and pushed. Verified the change on the browser

Phil 4.28.2022

Had some fun looking at trends based on the Twitter buyout:

“free speech” has an interesting trajectory over the last year:

Book

  • Cleaning up chapters

SBIRS

  • 9:15 standup
  • SimAccel meeting with Rukan – got things set up in the right way, I think
  • SimAccel API project. Ron has set up
  • RCSNN, hopefully. Yes! Set up calling methods in a class, which meant refactoring a bit so that there is now a BaseBoardMonitor class. More tomorrow.

GPT Agents

  • Still didn’t get to the IRB stuff

Phil 4.27.2022

https://faculty.cc.gatech.edu/~srijan/pubs/conflict-paper-www18.pdf

Get new UMBC ID! It’s ready!

SBIRs

  • 9:00 LM meeting
  • 1:00 DSCC Kickoff
  • Start on RCSNN story for this sprint
  • Some cool GPT debugging(?):
https://twitter.com/megamor2/status/1519291039479214080

Antigenic cartography has its roots in a mathematical technique called “multidimensional scaling,” which has been around since the 1960s. The algorithm uses data about the distances between pairs of objects to reconstruct a map of the objects’ relative locations. For example, if you had a table that lists the distances between a bunch of U.S. cities—like you might find in a road atlas—you could use a multidimensional scaling algorithm to reconstruct a map of those cities based solely on the distances between them. (IEEE Spectrum – The Algorithm that Mapped Omicron Shows a Path Forward)

Book

  • Read the epilogue to Aaron last night and made some tweaks. I need to work on the suggestions
  • Got a firm “no” and no leads from Kendall Hunt. Sigh

GPT Agents

  • Need to do informed consent, recruiting flyers and emails

Phil 4.26.2022

Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer

  • Hyperparameter (HP) tuning in deep learning is an expensive process, prohibitively so for neural networks (NNs) with billions of parameters. We show that, in the recently discovered Maximal Update Parametrization (muP), many optimal HPs remain stable even as model size changes. This leads to a new HP tuning paradigm we call muTransfer: parametrize the target model in muP, tune the HP indirectly on a smaller model, and zero-shot transfer them to the full-sized model, i.e., without directly tuning the latter at all. We verify muTransfer on Transformer and ResNet. For example, 1) by transferring pretraining HPs from a model of 13M parameters, we outperform published numbers of BERT-large (350M parameters), with a total tuning cost equivalent to pretraining BERT-large once; 2) by transferring from 40M parameters, we outperform published numbers of the 6.7B GPT-3 model, with tuning cost only 7% of total pretraining cost. A Pytorch implementation of our technique can be found at this http URL and installable via `pip install mup`.

sktime

sktime features a unified interface for multiple time series learning tasks. Currently, we support forecastingtime series classification and time series regression. We have experimental support for time series clustering and time series annotation.

Features:

  • API for machine learning with time series, for the purpose of specifying, fitting, applying and validating machine learning models
  • Interactive user experience with scikit-learn like syntax conventions

Book

  • More epilogue
  • Chase TODOs?

SBIRs

  • 10:00 proposal meeting. Make slides on background and concept
  • Add story for SimAccel library productization
  • 2:00 Sprint planning

GPT Agents

  • Put together study documents
  • 3:30 Meeting

Phil 4.25.2022

I helped build ByteDance’s vast censorship machine

  • “It was certainly not a job I’d tell my friends and family about with pride. When they asked what I did at ByteDance, I usually told them I deleted posts (删帖). Some of my friends would say, “Now I know who gutted my account.” The tools I helped create can also help fight dangers like fake news. But in China, one primary function of these technologies is to censor speech and erase collective memories of major events, however infrequently this function gets used.”

SBIRs

  • 9:00 Sprint Demos
  • Put together stories for next sprint
  • 2:00 MDA meeting
  • 4:00 OSS Meeting

Book

  • Reworked the epilogue a lot

Phil 4.22.2022

POT: Python Optimal Transport

SBIRs

  • Prep for 2:00 meeting

Phil 4.21.2022

Book

  • Finished first pass at GPT-3 interview
  • Work on Epilogue. Done? At least the first pass
  • Start hunting down TODO’s

SBIR’s

  • 9:15 Standup
  • Write up the current RCSNN results and create slides for tomorrow’s presentation with Rukan
  • I think that the error calculation should be a statistical measure of the divergence over a given window, which could include the entire prediction, but would otherwise be a tail of n
    • The average error
    • The variance (std dev, etc)
    • Outliers (e.g. rare, BIG errors)
    • Linear regression calc for error and variance
  • The reason that I think that this matters is that a prediction could have occasional large errors but otherwise be good, and we need a way to know that and characterize our results

GPT Agents

  • Getting back up to speed on Jarod’s work, which looks amazing.
  • Need to find some synonyms for slur

Phil 4.20.2022

Planting Undetectable Backdoors in Machine Learning Models

  • Given the computational cost and technical expertise required to train machine learning models, users may delegate the task of learning to a service provider. We show how a malicious learner can plant an undetectable backdoor into a classifier. On the surface, such a backdoored classifier behaves normally, but in reality, the learner maintains a mechanism for changing the classification of any input, with only a slight perturbation. Importantly, without the appropriate “backdoor key”, the mechanism is hidden and cannot be detected by any computationally-bounded observer. We demonstrate two frameworks for planting undetectable backdoors, with incomparable guarantees.
  • First, we show how to plant a backdoor in any model, using digital signature schemes. The construction guarantees that given black-box access to the original model and the backdoored version, it is computationally infeasible to find even a single input where they differ. This property implies that the backdoored model has generalization error comparable with the original model. Second, we demonstrate how to insert undetectable backdoors in models trained using the Random Fourier Features (RFF) learning paradigm or in Random ReLU networks. In this construction, undetectability holds against powerful white-box distinguishers: given a complete description of the network and the training data, no efficient distinguisher can guess whether the model is “clean” or contains a backdoor.
  • Our construction of undetectable backdoors also sheds light on the related issue of robustness to adversarial examples. In particular, our construction can produce a classifier that is indistinguishable from an “adversarially robust” classifier, but where every input has an adversarial example! In summary, the existence of undetectable backdoors represent a significant theoretical roadblock to certifying adversarial robustness.

Book

  • Work on the interview section. Ask about forms of bias, and how using the machine to find bias could help uncover patterns of it in humans as well. The idea of asking the same question a thousand times and getting a distribution of answers. Done! At least the first draft
  • Add something to the Epilogue about the tension between authoritarian and egalitarian governments
  • Play around with titles

SBIRs

  • 9:00 ITM discussion
  • Continue code generation. Need to make the BoardMonitor and BoardMonitorChild classes, then start running/stepping code within tool. I’d like to figure out tabs so that the JSON and hierarchy views could share the same screen space. Done!
Progress!
  • And remarkably, everything still works. Need to wire up the output of the dictionary

GPT Agents

  • Make a flier, email, and informed consent
  • Poke around at getting more technical keywords for things like science papers

Phil 4.19.2022

https://arxiv.org/abs/2203.11370

Language modeling via stochastic processes

  • Modern language models can generate high-quality short texts. However, they often meander or are incoherent when generating longer texts. These issues arise from the next-token-only language modeling objective. To address these issues, we introduce Time Control (TC), a language model that implicitly plans via a latent stochastic process. TC does this by learning a representation which maps the dynamics of how text changes in a document to the dynamics of a stochastic process of interest. Using this representation, the language model can generate text by first implicitly generating a document plan via a stochastic process, and then generating text that is consistent with this latent plan. Compared to domain-specific methods and fine-tuning GPT2 across a variety of text domains, TC improves performance on text infilling and discourse coherence. On long text generation settings, TC preserves the text structure both in terms of ordering (up to +40% better) and text length consistency (up to +17% better). Human evaluators also prefer TC’s output 28.6% more than the baselines.

Book

  • Finished(?) definitions
  • Moved “Interview with a Biased Machine” to the beginning of the Practice section. Going to work on that next

SBIRs

  • Get the lit review slides together for after the standup – done!
  • 9:15 standup
  • More code generation
    • Finish breaking bdmon into a class. As I do this, I think that it might make sense to have two directories – the directory that contains the editable child classes and a directory under that one that contains the generated files that are created each time the tool runs. Done!. This would allow the BoardMonitor class to have a decision_process() method that gets overridden easily in a child class. Next.
    • Dynamically calculate the import lib.
    • Wire up the run and step buttons
    • Terminate() should write things out? Done
  • Meeting with Ron about Crossentropy

GPT Agents

  • Figured out how to start find the Kuali IRB process and got some things down. Will need to walk through some things at the 3:30

Phil 4.18.2022

SBIRs

  • Lit review – goal is high quality and relevant
    • Two purposes – understanding the SoA, and finding a gap. This requires critical thinking, and an understanding of the problems, not just appeals to authority
      • Truthiness != trustworthiness
    • Wikipedia, Google, GScholar, and Elicit
      • Also blog posts, videos, etc.
    • Look at cites. Large counts are good! Search within citing
    • Look at authors. Sort by date. Is this recent?
    • Look for survey papers
    • Finding terms to search on is hard. Do not assume that you have the right ones at first.
    • Language model networks
  • Code generation
    • The subclassed code works!
    • Working on executing Python within python. It’s surprisingly easy. You can import the file/class, and then refer to it:
    def run_code_callback(self):
        self.dp.dprint("Run code")
        bdm = importlib.import_module("rcsnn.generated.bd_mon")
        bdm.main()

Phil 4.15.2022

Semantic Exploration from Language Abstractions and Pretrained Representations

  • Continuous first-person 3D environments pose unique exploration challenges to reinforcement learning (RL) agents because of their high-dimensional state and action spaces. These challenges can be ameliorated by using semantically meaningful state abstractions to define novelty for exploration. We propose that learned representations shaped by natural language provide exactly this form of abstraction. In particular, we show that vision-language representations, when pretrained on image captioning datasets sampled from the internet, can drive meaningful, task-relevant exploration and improve performance on 3D simulated environments. We also characterize why and how language provides useful abstractions for exploration by comparing the impacts of using representations from a pretrained model, a language oracle, and several ablations. We demonstrate the benefits of our approach in two very different task domains — one that stresses the identification and manipulation of everyday objects, and one that requires navigational exploration in an expansive world — as well as two popular deep RL algorithms: Impala and R2D2. Our results suggest that using language-shaped representations could improve exploration for various algorithms and agents in challenging environments.

Tasks

  • Mulch and edging
  • Fortunately, taxes are already done!
  • Maybe get started on chores

SBIRs

  • Send text to JHU – done! But they aren’t going for it
  • Code generation
    • Made some buttons that trigger non-functional callbacks
    • Got the immutable-ish child classes working

GPT Agents

  • Upload Yelp paper to ArXiv – done!

Book

  • Start finishing deep bias – done?
  • Definitions

Ending the week with this:

Phil 4.14.2022

https://twitter.com/francoisfleuret/status/1514684663310295041

Tasks

  • Mulch and edging

Book

  • Sent proposal to KH
  • Finished Hierarchy in the Forest. I need to scan the marked pages
  • Still need to finish the Deep Bias chapter

SBIRs

  • 9:15 Standup
  • Follow up with Rukan about entropy and accumulated error
  • Meeting with Ron about GPT
  • See if there is general interest in lit review tools – Yes set something up for next week
  • Code generator

GPT Agents

  • Try to set up IRB submission?

Phil 4.13.2022

Book

  • Moved some text around to the beginning GPT interview and took it out of the influence/dominance/attention section. I had to rework that a bit to include egalitarianism and inverse dominance
  • Trying to figure out how to finish up the deep bias chapter. I’d like to do something that shows how these patterns play out in modern politics. Maybe the difference between suppression and cancelling
  • 1:00 Meeting! It went well, I think. KH is a textbook company, so it’s probably not a good fit but 1) I found a way to talk to publishers! and 2) They will take a look at the proposal and make suggestions (maybe?)

SBIRs

  • 8:30 IRAD meeting
  • 10:00 LM Catching up
  • 11:00 Goals
  • Finish goals (add measures)
  • Write abstract – done
  • Write one-pager for Dave M. – done
  • Work on code generator – nope

Phil 4.12.2022

Book

  • Starting to finish up Deep Bias chapter. Maybe move it to the front? My thinking is to introduce the human tension between hierarchy and egalitarianism, then communication technology (phase locking), then iteratively revisit?

SBIRs

  • Meeting with Steve – done
  • Sprint planning – done
  • Write up notes from yesterday
  • Set up MDA meeting for next week?

GPT-Agents

  • 3:30 Meeting
  • Since ASRC is unwilling to be lead, do we write a proposal? Find a lead?