AUTONOMOUS-FOX
Three streams of interconnected research. One objective: a general-purpose model of market behaviour.
Three interconnected research streams. One objective: a general-purpose model of market behaviour.
We develop fundamental machine-learning methods for learning the structure of complex systems, with financial markets as our focus. The challenge extends beyond modelling individual assets: it involves learning how behaviour develops through time and how changing interactions produce structure across the wider market.
Our Expertise
Stream 1: Learning the Building Blocks of Market Behaviour
Raw market observations mix useful information with noise. Similar behaviour can appear in different forms, while superficially similar movements can reflect different conditions. Learning a useful representation requires identifying which distinctions matter and which variations obscure them.
Our first research stream develops methods for learning behavioural building blocks directly from data, without defining them in advance. The vocabulary analogy is useful: these building blocks provide a basis for learning more complex organisation. Our research concerns what they capture, how faithfully they preserve information and how learned abstractions can support modelling across different contexts.
Stream 2: Discovering How Markets Change
A pattern observed at one moment tells only part of the story. Its significance depends on what preceded it, the timescales over which it developed and the context in which it occurs. Understanding market behaviour requires learning these dependencies as well as recognising individual patterns.
Our second research stream develops methods for learning temporal structure from market representations. We investigate how patterns combine into sequences, how dependencies extend across timescales and how their evolution changes with context. The objective is a representation of market dynamics that different downstream models can use for their own purposes.
Stream 3: Mapping the Connections Between Assets
Markets are complex systems whose behaviour emerges through interactions. Relationships form within and between groups of assets, overlap across scales and change through time. Capturing this organisation requires more than a fixed map of pairwise connections.
Our third research stream advances relational structure learning, drawing on geometric deep learning and investigating richer forms of hierarchical organisation. We develop methods to represent interactions among assets and groups, and to learn how those structures evolve alongside individual behaviour. This is frontier research into how AI can represent the organisation of a changing system.
How They Fit Together
The three streams address connected aspects of the same scientific problem: what to represent, how it evolves and how it interacts. We develop them as a coordinated research programme because the useful information in each depends on the others.
Our objective is a general-purpose representation that preserves these dependencies and makes them available to downstream models. The research concerns both the individual methods and how they work together to capture structure that would otherwise remain difficult to learn from raw observations.
A Foundation for Multiple Strategies
The same market representation could inform multiple trading strategies simultaneously. Each strategy would retain its own signals, objectives and decisions while drawing on information about the wider system that might otherwise fall outside its scope.
Execution, portfolio construction, risk management and market making offer further applications of the same research. Our focus is the scientific foundation: developing representations whose usefulness extends beyond any single trading task.