Antigenic cartographyhas 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)
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
Need to do informed consent, recruiting flyers and emails
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`.
“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.”
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.
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
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!
And remarkably, everything still works. Need to wire up the output of the dictionary
Make a flier, email, and informed consent
Poke around at getting more technical keywords for things like science papers
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.
Moved “Interview with a Biased Machine” to the beginning of the Practice section. Going to work on that next
Get the lit review slides together for after the standup – done!
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
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
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.
Mulch and edging
Fortunately, taxes are already done!
Maybe get started on chores
Send text to JHU – done! But they aren’t going for it
Made some buttons that trigger non-functional callbacks
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?)
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?
Meeting with Steve – done
Sprint planning – done
Write up notes from yesterday
Set up MDA meeting for next week?
Since ASRC is unwilling to be lead, do we write a proposal? Find a lead?