Next was a great talk by @jascha on learned optimizers at #IAIFI. This work is going after the important problem of moving away from hand-designed optimizers in deep learning, and Sohl-Dickstein shows some promising results here https://www.youtube.com/watch?v=FrqLLRpAdL0 (4/11)

📢New paper (finally...) out today📢 'EPiC-GAN: Equivariant Point Cloud Generation for Particle Jets' with Erik (#UniHamburg too) and Jesse (#IAIFI)

A short summary below, full paper at https://arxiv.org/abs/2301.08128

One useful way to represent data from particle physics collisions is a point cloud: each collision event is a cloud of points & each point has a position in space (the position of the specific sensor or particle) and some additional features attached (for example the energy)