Writing
Learning in public
Shipping what I learn.
Latest posts 6 posts
Extending unitary convolutions for learning on graphs
Reducing oversmoothing with unitary graph convolutions.
Still lost in the middle
Why models forget the middle of a prompt.
Even more efficient finetuning
Fewer trainable parameters, better performance.
Optimization by prompting
Prompting as a lightweight optimization tool.
More efficient finetuning
Low-rank adapters for fast, local finetuning.
Transformers: unpacking the buzzword
What are transformers?