Luka Ribar Email & Phone Number

London, England, United Kingdom

My research mainly focuses on developing training and inference algorithms for large language models. I am particularly interested in efficient techniques leveraging different forms of sparsity and low precision. Projects:• Training compute-efficient models using sparsely-activated mixture-of-experts layers.• Sparse attention techniques for fast autoregressive transformer inference.• Quantisation and low-precision numerical formats for model compression.Publications:• SparQ Attention: Bandwidth-Efficient LLM Inference (ICML 2024)

Cambridge, England, United Kingdom

• Developed system architecture and methods for mimicking robustness and reliability of neural systems, focusing on mechanisms of neuromodulation.• Handled different aspects of the project including dynamical systems modelling and analysis, developing the software environment and tools in Python and circuit design.

Cambridge, England, United Kingdom

• Supervised students in signal processing and control.• Supervised two master's research projects in model predictive control and neuromorphic engineering.

Petnica, Serbia

• Supervised and lectured high school students in physics and electronics courses.

Belgrade, Serbia

• Developed an optical character recognition system for cursive scripts using Hidden Markov Models.

Doctor Of Philosophy (Phd), Control And Neuromorphic Engineering

• Thesis title: "Synthesis of neuromorphic circuits with neuromodulatory properties" • Supervisor: Prof. Rodolphe Sepulchre • Applied.

Master Of Engineering (Meng), Electrical And Information Sciences, First Class Honours

Activities and Societies: Cambridge University Yugosphere Society• Obtained First Class Honours in every year of study. • Ranked 5th out.

Focus On Mathematics, Physics And Computer Science, 5.00/5.00