Nathan Gavenski

O mundo não dá voltas, ele capota!😂 reporterbrasil.org.br/2025/01/venda-…

I don't know why, but having a masterclass from @JamieBGall on the beach in Brazil feels better than in London 😂😂😂

Made a watcher so I don't have to run my script manually when coding github.com/NathanGavenski… Discussions: discu.eu/q/https://gith… #programming #python by @NGavenski

Set your reminders for the 2024 #XmasLectures, supported by @CGI_global! 🗓️ 29, 30 and 31 December 📺 BBC Four and iPlayer ⏰ 9pm In the series, @DoctorChrisVT will explore the surprising science of food – from production to digestion. Find out more: rigb.org/christmas-lect…

Couldn't find the original paper from this, well... back to writing.

It got better

I'm always curious about what data goes into @GitHubCopilot. Today, when writing LaTeX locally, I decided to try it. My paper has nothing to do with it, so I guess papers also go into it. Nice try Copilot, nice try.

If you are interested in something that allows you to run scripts automagically after modifying files, here is a package you might enjoy! New release: github.com/NathanGavenski…

Having to work extra hard to surpass your own method should not be part of writing a new paper 😅

Me: I can finally submit my literature review to a journal. Journal: ... Me:

I couldn't agree more. A supportive supervisor has greatly elevated my work and myself as a researcher in the past few years. It is night and day!

Had an amazing time at PhD Connect 2024 in Leeds, and many thanks to @turinginst for organising the event. It is always an excellent opportunity to get to know many other PhD Students and discuss different research topics!

Yesterday I had the pleasure to present my work at @ecai2024. Thanks to all the authors for the great work, @safe_trusted_ai for supporting me and allowing me to pursue my research and @KingsNMES for the resources and space that makes this work possible.

Thanks to all the authors that made this work possible @odinaldo @drmeneguzzi @michaelluck and Juarez Monteiro

1/2 We use exploration and path signatures to create expert-like agents few expert samples and no prior knowledge/human intervention. You can read more about it in: ebooks.iospress.nl/doi/10.3233/FA… And check our public code in: github.com/NathanGavenski…

1/2 If you are interested in imitation learning and you are @ecai2024, come join me in my talk about our work "Explorative Imitation Learning: A Path Signature Approach for Continuous Environments" I will talk as part of the machine learning 2 session tomorrow (21/10)

1/5 Suppose you need 40% of a model's trained parameters to keep 95% of the model’s accuracy What proportion of parameters do you need if you make the model 5x wider? 🤔 Come to tomorrow’s @ecai2024 presentations to find out!

@Michael_J_Black The review process needs a remake. Reviewing 6-8 papers for conferences on topics that aren't close to your research gives the right excuse to reviewers rely on LLMs. I recently received my first LLM review and it was nonsensical, and the reviewer rejected our work with no basis.

AI-generated reviews of scientific papers are increasing, vacuous, and need to be stopped quickly. They reduce author trust in the review process. Proposal: someone who is judged to have submitted such a review is banned from submitting to the same conference/journal for 2 years.

Here is something I didn't know I needed 😂 next paper for sure will have loss rider graphs

It was a great experience to go to AAMAS and get to know so many researchers and such a beautiful place! High expectations for @ecai2024, and thrilled to present our current work there.