Research

Predicting the future is a theoretical problem before it's a modelling one. Our research looks for new ways to resolve outcomes, elicit private information, and turn both into signal a model can learn from.

Research focus

Algorithmic economics

  • Abstract visual for prediction marketsPrediction MarketsMarkets that price the likelihood of future events, and the scoring rules that keep those prices honest.
  • Abstract visual for forecastingForecastingEliciting and aggregating dispersed beliefs into calibrated predictions that hold up against outcomes.
  • Abstract visual for cryptoeconomicsCryptoeconomicsIncentives that make decentralised systems behave, so participation and honesty pay better than manipulation.

Machine learning

  • Abstract visual for distributed learningDistributed LearningTraining and inference spread across heterogeneous machines, with communication as the binding constraint.
  • Abstract visual for security of machine learningSecurity of Machine LearningAttack surfaces that open up once training leaves a trusted datacentre, and the defences that close them.
  • Abstract visual for verifiable learningVerifiable LearningProving a model ran as claimed, so results from an untrusted machine can still be trusted.

Our research team

Recent research

Research
IR3DE-AXL: Collective Inference Network
Research
Credibly Neutral AI Oracles
Research
The Role of Feedback Alignment in Self-Distillation
Research
Markets That Explain Themselves
Research
Look Beyond One-Size-Fits-All LLMs with IR3DE
Research
DEI: Diversity in Evolutionary Inference for Quality-Diversity Search
Research
F-TIS: Heterogeneous GRPO Without Homogeneous Assumptions
Research
Backdoor in the Middle: Attacking Pipeline Parallelism
Research
Stop Multitask Training. Just DUME.