Adversarial attacks can completely screw up an image recognition model's performance - without humans noticing a single difference. What are these attacks and how can we defend against them? - we'll explore papers and perspectives on this topic in this episode. View article summaries and links at the website here: http://papers-on-the-go.ml --- Send in a voice message: https://podcasters.spotify.com/pod/show/andre-ye/message

How can a completely random neural network attain high performance? The answer lies in a very fascinating hypothesis - the Lottery Ticket Hypothesis - and allows us to make some very intriguing paradigm shifts in how we view neural network generalization. View article summaries and links at the website here: http://papers-on-the-go.ml --- Send in a voice message: https://podcasters.spotify.com/pod/show/andre-ye/message

Why don't neural networks overfit when they have so many parameters? It's deep learning's million-dollar question and one we'll try to get a few steps closer to answering. We'll get an introduction to the million-dollar question, get an understanding for the surprising amount of information neural networks can carry, and explore the discussion and research around Deep Double Descent phenomena. Links to the papers discussed: Understanding Deep Learning Requires Rethinking Generalization: arxiv.org/pdf/1611.03530.pdf || Reconciling modern machine learning practice and the bias-variance trade-off: arxiv.org/pdf/1812.11118.pdf || Scaling description of generalization with number of parameters in deep learning: arxiv.org/pdf/1901.01608.pdf Reach out to me at my email, [email protected], or at my LinkedIn, www.linkedin.com/in/andre-ye/. --- Send in a voice message: https://podcasters.spotify.com/pod/show/andre-ye/message

Hey everyone! I'm your host, Andre Ye. I’m super excited to be starting this new podcast, ML & DS Papers on the go. It is my hope that this podcast is able to bring the most fascinating and recent advancements in machine learning and data science to your ears for easy access. I hope you'll join me in future episodes! --- Send in a voice message: https://podcasters.spotify.com/pod/show/andre-ye/message