Technology
On Weak Regret Analysis for Dueling Bandits
When the optimality gap is negligible, we propose another algorithm that outperforms our first algorithm, highlighting the subtlety of this dueling bandit problem.
Parameters as interacting particles: long time convergence and asymptotic error scaling of neural networks
Grant Rotskoff, Eric Vanden-Eijnden
Theperformance ofneural networksonhigh-dimensional datadistributions suggests that it may be possible to parameterize a representation of agiven highdimensional function with controllably small errors, potentially outperforming standard interpolation methods. We demonstrate, both theoretically and numerically, that this is indeed the case. We map the parameters of a neural network to a system of particles relaxing with an interaction potential determined by the lossfunction.
On Making Stochastic Classifiers Deterministic
Andrew Cotter, Maya Gupta, Harikrishna Narasimhan
Neural Information Processing Systems http://nips.cc/
When to Trust Your Model: Model-Based Policy Optimization
Michael Janner, Justin Fu, Marvin Zhang, Sergey Levine
Then, at policy fork steps.
Cost(Q,G,F) = MX
Todetermine the best threshold, we first determine the value ofT that results inthe best AI-alone accuracy on a small subset of the ImageNet-ReaL (2K images) and CUB (1K images) datasets, and then we evaluate the AI-alone accuracy on the held-out set for each dataset (42K images on ImageNet-ReaLand4KonCUB).
basedexplanationsimproveA I
Our explanations help users more accurately reject AI'swrong decisions than all other tested methods.
