Domenic Rosati

About Me

I’m a computer science PhD student at Dalhousie University supervised by Hassan Sajjad and Frank Rudzicz.

My primary research interest is the safety of machine learning systems. At the moment, my main work is:

• Formulating safety as a learning theory problem
• Understanding how inference-time behaviour is connected to training dynamics
• Using safety engineering techniques like system scoping, specs, learning constraints, behavioural constraints, etc.
• Developing technical mechanisms with AI safety regulators and policy people
• Applying the above to empirical settings that look similar to very dangerous systems (agentic, mislaigned, adversarial, etc.)

If you want to work on these problems together let me know: domenic (dot) rosati (at) dal (dot) ca.

You can see work I am mostly finished with on my google scholar.

If you are looking for my career history in industry you can see here: https://www.linkedin.com/in/cinemod-itasor/

I am also interested in the following things that you can talk to me about:

• Chiptune
• Anime, TV, Movies (see this cool cyberpunk letterboxd)
• History of Ideas (particularly of math and cs for example stuff in Historia Mathematica - taking Focauldian/lingustic history recs)
• Science Fiction (I really really like short stories - please email me with good recommends)
• Philosophy of Science and Math (Lakatos, Popper)
• Applied Math
• Linguistics
• Dath Ilan
• Djent
• Power Lifting

Teaching & Learning

There is nothing like helping someone understand the Church–Turing thesis for the first time.

I like teaching machine learning and CS (I taught CS in High school for four years). Here is a collection of course notes and lectures I try to keep up to date.

Courses I’d like to teach AI Safety & Alignment, ML/DL/Statistical Learning Theory, History of Artificial Intelligence (or just history of modern mathematics), Computational Linguistics, Research Methods & Experimental Design, Optimization