the gears to ascension comments on the gears to ascenscion’s Shortform

the gears to ascension 13 Sep 2022 8:13 UTC
1 point
interesting capabilities tidbits I ran across today:
- 1: geometric machine learning and neuroscience: https://github.com/neurreps/awesome-neural-geometry
- 2: lecture and discussion links about bayesian deep learning https://twitter.com/FeiziSoheil/status/1569436048500920320
- 3: Learning with Differentiable Algorithms: https://twitter.com/FHKPetersen/status/1568310569148506114 - https://arxiv.org/abs/2209.00616
1: first paragraph inline:

A curated collection of resources and research related to the geometry of representations in the brain, deep networks, and beyond, collaboratively generated on the Symmetry and Geometry in Neural Representations Slack Workspace.

2: tweet thread inline:

“Myths and reality in Bayesian Deep Learning”, a lecture by Prof. Andrew Wilson in our deep learning foundations series: https://youtu.be/lhwk4ESlyMA Related papers: (1) https://arxiv.org/abs/2002.08791 - Bayesian Deep Learning and a Probabilistic Perspective of Generalization (2) https://arxiv.org/abs/2003.02139 - Rethinking Parameter Counting in Deep Models: Effective Dimensionality Revisited (3) https://arxiv.org/abs/2106.11905 - Dangers of Bayesian Model Averaging under Covariate Shift (4) https://arxiv.org/abs/2202.11678 - Bayesian Model Selection, the Marginal Likelihood, and Generalization (5) https://arxiv.org/abs/2112.01388 - Residual Pathway Priors for Soft Equivariance Constraints As always, all materials are posted on the course webpage: http://cs.umd.edu/class/fall2022/cmsc828W/

You can access my own video lectures from last years in our Youtube playlist: https://www.youtube.com/playlist?list=PLHgjs9ncvHi80UCSlSvQe-TK_uOyDv_Jf

3: key tweet inline:

By making algorithms differentiable, we can integrate them end-to-end into neural network machine learning architectures. For example, we can continuously relax sorting (http://github.com/Felix-Petersen/diffsort) for learning to rank. [2/13]