Latent Space Reasoning

Presentation Link

The sixtieth MeetUp of the Machine Learning Singapore Group, was titled : "News from the Frontier" - reflecting the timing, being just after Google Cloud Next (in Las Vegas) and ICLR (the prestigous academic conference, held this year in Singapore).

My Presentation

My talk was titled "Latent Space Reasoning", and had the following outline:

• Latent Space Reasoning
  • What's involved?
  • Three different techniques
• ICLR
  • Three paper takeaways
• Head's Up!
  • AMD kernel competition; Qwen3; and ChatGPT Ads
• Wrap-up & QR-code (the latter to reduce audience distractions)

To set the scene, I lead the audience through the the information capacity of the token stream (only 18-bits per token), and compared that to the far larger latent space dimension. Then I covered three different kinds of Latent Space Reasoning:

• Recurrent Depth (Huginn)
• Latent Tokens (eg: COCONUT)
• Large Concept Models (from Meta's FAIR)

The slides for my talk, which contain links to all of the reference materials and sources, are here :

If there are any questions about the presentation please ask below, or contact me using the details given on the slides themselves.

Other Presentations

In his talk "Agent Announcements & Trends from Google Cloud Next 2025", Sam Witteveen explained the new Agent frameworks/stratgey launched by Google at the Cloud Next event in Las Vegas, and also described some of the approaches also being presented by Google partner companies there.

We were also very please to welcome two other speakers to describe their ICLR take-aways:

• Jenny Zhang, with "Open-endedness and Interestingness" - giving highlights from talks by Tim Rocktäschel & Jeff Clune (her PhD supervisor in British Columbia); and
• Raymond Chan, whose "ICLR Highlights" focussed on topics in safety and alignment of LLMs

Acknowledgements

Many thanks to the Google team, who not only allowed us to use Google's Developer Space, but were also kind enough to provide Pizza for the attendees!