Machine Learning Research Scientist

Key Responsibilities: Design and implement custom decoder-only Transformer architectures for financial time-series modelling. Develop tokenisation approaches for intraday market data, incorporating price movements, volume, order flow and cross-sectional features. Build efficient PyTorch training pipelines using mixed-precision training, gradient checkpointing and multi-GPU parallelism. Research attention mechanisms that capture temporal relationships, cross-asset dependencies and patterns across multiple timescales. Develop evaluation frameworks to assess predictive accuracy, signal quality and trading performance. Optimise model inference for low-latency production deployment through compression, quantisation and efficient decoding techniques. Conduct rigorous experiments and ablation studies to validate architectural choices and training methodologies. Collaborate with researchers and developers to integrate model predictions into live trading infrastructure. Define and execute a sustained research agenda, assessing new approaches and refining models through systematic experimentation. Document research methodology, experimental findings and architectural decisions.

Requirements: PhD in Machine Learning, Computer Science, Statistics, Applied Mathematics or a related discipline, with a focus on deep learning. Demonstrated experience implementing Transformer architectures from first principles. Deep understanding of attention mechanisms, positional encodings, tokenisation strategies and model training dynamics. Expert-level PyTorch skills, including custom modules, training loops, mixed-precision training and multi-GPU training. Experience training large-scale models with 100 million or more parameters. Strong mathematical foundations in linear algebra, probability, optimisation and information theory. Strong programming skills in Python and C++ for performance-critical components. Ability to independently define and execute a research agenda spanning several months. Familiarity with AI-assisted development tools. Experience applying deep learning to financial data or time-series forecasting, including tokenisation of continuous or non-text data, is advantageous. Published research at leading machine learning conferences, such as NeurIPS, ICML or ICLR, or equivalent industry research experience is preferred. Knowledge of market microstructure, intraday trading dynamics, model compression and inference optimisation is beneficial. For more information contact: Graham Murphy – graham@pointonetalent.com