Edwin V. Bonilla

“It’s not [only] the consequence that makes a problem important, it is that you have a reasonable attack.”
Richard Hamming, 1986.

I have been working on machine learning and AI for 20+ years and currently lead Foundational Machine Learning at CSIRO, where my focus is on making probabilistic machine learning a fundamental part of decision-making, under uncertainty, at scale, and grounded in causal understanding.

That’s meant building things, not just publishing about them: I pioneered multi-task Gaussian process methods now widely used for transfer learning under uncertainty; co-created Arrow, one of the first foundation models for causal discovery; and built platforms like AutoGP (scalable Gaussian process modelling) and VGCN (graph-structure learning) that turned research into reusable infrastructure. I have developed deep reinforcement learning approaches to train agentic policies for Bayesian sequential experimental design. I also led the ML system behind Milepost GCC, the first machine-learning-driven optimising compiler, still cited today as test-of-time work.

At CSIRO I lead a team of researchers and engineers, set technical strategy for causal AI, generative AI, and decision intelligence, building on transformer and LLM-based foundations, and translated that work into real solutions across renewable energy, climate systems, aerospace, and education. I’ve taken a research capability from zero to organisational strategy, and I care about doing that kind of work at scale.

My work is published regularly at NeurIPS, ICML and ICLR (6,000+ citations, a recent ICML oral, two test-of-time awards, a NeurIPS Graph Representational Learning Workshop Outstanding Contribution Award).