Can Latent Program Networks Solve Abstract Reasoning? [Clement Bonnet] @MachineLearningStreetTalk
Can Latent Program Networks Solve Abstract Reasoning? [Clement Bonnet]  @MachineLearningStreetTalk
Uploaded February 2025 | Updated November 2026, 4 days ago
SPONSOR MESSAGES:
***
CentML offers competitive pricing for GenAI model deployment, with flexible options to suit a wide range of models, from small to large-scale deployments. Check out their super fast DeepSeek R1 hosting!
centml.ai/pricing

Clement Bonnet presents Latent Program Networks (LPNs), a fundamentally different approach to the ARC-AGI benchmark that embeds programs into a continuous latent space rather than generating code or fine-tuning language models. The architecture uses a VAE-style encoder to map input-output pairs into distributional program representations, performs gradient-based search during both training and inference to refine these representations, and decodes solutions directly without an explicit program intermediary.

The conversation explores why ARC remains impervious to standard neural networks -- Chollet designed it to resist memorization, with test tasks far enough from any training distribution that zero-shot generalization fails. Bonnet and Tim Scarfe dig into the induction vs. transduction distinction, the role of the kernel trick in blurring that boundary, and why parameter-space search (test-time training) may be less efficient than searching a compressed program manifold.

A key architectural insight: training the search procedure end-to-end forces the latent space to become smooth and searchable, similar to MAML-style meta-learning. The system achieves ~10% on ARC evaluation sets using only small transformers (40M parameters total) trained from scratch on Michael Hodel's re-ARC dataset -- no pre-trained LLMs, no internet priors.

The discussion also probes the limits of continuous latent spaces for compositional reasoning. Bonnet acknowledges that single-thread search through a fixed latent space cannot achieve true program composition, but suggests multi-threaded search with external composition could bridge the gap. The episode closes with reflections on creativity, the efficiency gap between human hypothesis-testing (a handful of guesses) and LLM program sampling (millions of samples), and what scaling LPNs might reveal about the structure of program spaces.

---
REFERENCES:
paper:
[00:00:05] ARC-AGI Benchmark
arxiv.org/abs/2412.04604
[00:02:10] Latent Program Networks
arxiv.org/abs/2411.08706
[00:08:45] Induction vs Transduction in Abstract Reasoning
arxiv.org/abs/2411.02272
[00:17:40] Variational Autoencoders
arxiv.org/abs/1312.6114
[00:33:00] Critique of Deep Learning
arxiv.org/abs/2002.06177
person:
[00:07:45] Kevin Ellis - Program Synthesis
https://www.cs.cornell.edu/~ellisk/
tool:
[00:27:50] re-ARC Framework
github.com/michaelhodel/re-arc

---
LINKS:
Full Transcript: app.rescript.info/share/55768bb71fbcc62c0522136650ccf960
Download PDF transcript: app.rescript.info/api/public/sessions/54ec5e9b3152760b/pdf

Clem and Matthew-
linkedin.com/in/clement-bonnet16
github.com/clement-bonnet
mvmacfarlane.github.io

[00:27:50] re-ARC, Hodel
github.com/michaelhodel/re-arc

[00:29:40] Grid size in ARC tasks, Chollet
github.com/fchollet/ARC-AGI
Can Latent Program Networks Solve Abstract Reasoning? [Clement Bonnet]Mutually Assured AI Malfunction [Dan Hendrycks]Your Brain Is a Prediction Machine, Not a Processor — Karl FristonChatGPT is overconfident (Anil Ananthaswamy)People wanted faster horses - Jeff CluneThe Mathematical Foundations of Intelligence [Professor Yi Ma]Why Humans Are Still Powering AI [Sponsored] - Phelim BradleyWhen We Asked GPT-4 to Solve a Problem, It Chose Blackmail — Sara Saab & Enzo BlindowOptimize GPU performance for AI - Prof. Gennady PekhimenkoYour Brain Doesnt Command Your Body. It Predicts It. [Max Bennett]Impostor Intelligence?Build Specialist LLMs Like It’s 2019 (Randall Balestriero)
Machine Learning Street Talk |

Can Latent Program Networks Solve Abstract Reasoning? [Clement Bonnet]

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER