Joseph Viviano

lots of people offering jobs over there

but conversation is much better over here

get grinding with claude asap, build that muscle

We need to raise the bar on research code right now.

1) documentation and tests are dead simple now.
2) creating benchmarks integrating across multiple implementations
3) have agents double check your work / fix broken tests
4) fix outstanding bugs in major scientific packages

GitHub - utilForever/baba-is-auto: Baba Is You simulator using C++ with some reinforcement learning Baba Is You simulator using C++ with some reinforcement learning - utilForever/baba-is-auto

@hempuli.bsky.social !

We want to expand on utilForever's Baba is You RL simulator!

github.com/utilForever/...

Is that ok with you? We want to expand the set of levels included to involve more complex problems and reasoning over the various abstractions in your (awesome and very hard) game!

Salem Lahlou Archives Assistant Professor of Machine Learning

Massive shoutout to the efforts of Sanghyeok Choi (he's on the dark MAGA app), Salem Lahlou (mbzuai.ac.ae/study/facult...), and ‪‪@oyounis.bsky.social - this was very much a team effort and we’re really excited to help popularize gflownet use through these tools. We really value your feedback!

If you’re interested in using torchgfn, helping us improve the library, want help incorporating torchgfn into your workflow, or have any feedback, please feel free to familiarize yourself with our documentation and reach out – there’s still lots to do!

torchgfn.readthedocs.io/en/latest/

Moving forward, we plan to focus on optimizing the library for large-scale distributed training setups, and supporting more specialized and demanding environments, particularly in the AI for Science domains.

The structure of the GFlowNet itself is highly modular, permitting the use of modified losses, custom samplers, novel off-policy sampling methods, and new policy architectures with minimal changes to the underlying library elements.

Basic torchgfn usage follows standard pytorch workflows, allowing the user to swap in any modified components to support the development of new methods:

This is a major update:

+ Much easier environment definition.
+ Cleaner abstractions → easier extensibility!
+ Support for graph-based states under torch_geometric.
+ Improvements to every core element of the library.
+ Lots of new environments, tutorials, and examples!

GitHub - GFNOrg/torchgfn: A modular, easy to extend GFlowNet library A modular, easy to extend GFlowNet library. Contribute to GFNOrg/torchgfn development by creating an account on GitHub.

We’ve released torchgfn v2!

github.com/GFNOrg/torch...

We believe we’ve built the go-to library for fundamental GFlowNet methods development and prototyping, and we’re really excited to help you start using it.

Today marks a big milestone for me. I'm launching @law-zero.bsky.social, a nonprofit focusing on a new safe-by-design approach to AI that could both accelerate scientific discovery and provide a safeguard against the dangers of agentic AI.

Yeah the lack of memepoasters and tpot adjacent attention bait makes the platform great for us but bad for it taking over the disinformation psyop town square.

America cannot long remain free, nor first among nations, if it becomes the kind of place where universities are dismantled because they don't align politically with the current head of the government.

Preprint Alert 🚀

Can we simultaneously learn transformation-invariant and transformation-equivariant representations with self-supervised learning?

TL;DR Yes! This is possible via simple predictive learning & architectural inductive biases – without extra loss terms and predictors!

🧵 (1/10)

Great job gang! And thanks to @boussifo.bsky.social for being such a stellar lead. (Also depicted: @jainmoksh.bsky.social)

Action abstractions for amortized sampling As trajectories sampled by policies used by reinforcement learning (RL) and generative flow networks (GFlowNets) grow longer, credit assignment and exploration become more challenging, and the long pl...

@boussifo.bsky.social , Léna Néhale Ezzine, Michał Koziarski, @jainmoksh.bsky.social , Nikolay Malkin, @bengioe.bsky.social , Rim Assouel, @yoshuabengio.bsky.social - @mila-quebec.bsky.social

Action Abstractions for Amortized Sampling
📄 arxiv.org/abs/2410.15184
💻 github.com/GFNOrg/Chunk...

The additive benefit of combining chunking with diversity-seeking samplers, like GFlowNets, also points towards an intriguing explanation as to why macro action discovery has not been found generally useful in the RL context.

💡 Why is this exciting?

Hierarchical planning is a key component of intelligence—both biological & artificial. By dynamically learning & using abstractions, our method bridges the gap between RL, program induction, and cognitive science.

These chunks also generalize - they’re transferable across samplers and tasks!

Chunks learned in one environment improve exploration and sampling efficiency in unseen settings, suggesting the method abstracts high order general principles that are robust & adaptable to new envs!

For mode discovery, our approach also significantly speeds up discovering diverse high-reward states.

For example, in FractalGrid, vanilla GFlowNets get stuck in a single mode, but armed with ActionPiece, it unlocks new exploration paths!

Chunking helps!

Across synthetic and real-world tasks (e.g., RNA sequence generation, bit sequences, and FractalGrid), our approach improves especially for GFlowNets:

✅ Mode discovery
✅ Exploration
✅ Density estimation
✅ Interpretability

By applying BPE (which we're calling "ActionPiece" for learning chunked actions) to sampled trajectories, we extract meaningful high-level actions that naturally emerge during learning. For example, here are some learned chunks from sampler of RNA binders:

We chunk frequently occurring subsequences into high-order actions using Byte Pair Encoding (BPE)—a popular NLP tokenization technique. These chunks are added to the action space, which progressively reduces trajectory length and helps uncover latent structures in the task.

In RL & GFlowNets, with longer trajectories, assigning credit and discovering diverse high-reward states gets harder. Standard methods struggle to sample structured distributions efficiently & many previous attempts to discover high-order actions failed to show consistent benefit.

Ecstatic to show off some work my brilliant colleagues and I did at @iclr-conf.bsky.social this year! 🚀

We address the credit assignment challenge under long trajectories in RL or GFlowNets by constructing high order actions, or “chunks”, effectively compressing trajectory lengths!

From having to wrap a bunch of methods I can say I'm not a big fan of having configs in the form of argparsers.

Musk tweet: CFPB RIP

any economists here that can help me understand why the world's richest man might want to kill the consumer financial protection bureau

Neural surrogates can accelerate PDE solving but need expensive ground-truth training data. Can we reduce the training data size with active learning (AL)? In our NeurIPS D3S3 poster, we introduce AL4PDE, an extensible AL benchmark for autoregressive neural PDE solvers. 🧵

UnitedHealth faces class action lawsuit over algorithmic care denials in Medicare Advantage plans A class action lawsuit was filed against UnitedHealth and a subsidiary alleging they use an algorithm to deny care to seriously ill patients

"The complaint alleges the algorithm, dubbed nH Predict, has a 90% error rate, basing that calculation on the percentage of payment denials reversed through internal appeals processes or administrative law judge rulings."

www.statnews.com/2023/11/14/u...