MSc in AI Systems from the University of Trento. I work on machine learning and NLP, and I'm fascinated by the messier corners of it: bias in generative models, and how ideas from neuroscience end up intertwining with AI
Thesis: What are we really decoding in musical attention? It started as an attempt to improve an existing EEG-audio decoding architecture (Akama et al.) and turned into an investigation of what music attention-decoding models actually measure. Code and evidence are public.
The pinned repos below may be the ones worth your time: the thesis, ProtoCoCoOp (adapting CLIP without losing zero-shot), the NLU project (a language model plus intent and slot filling on ATIS), and OpenInterBias.
Reach me at andrea.giampietro@proton.me or on LinkedIn.