no code implementations • 16 Jun 2020 • Jñani Crawford, Eshed Margalit, Kalanit Grill-Spector, Sonia Poltoratski
To be interpretable and useful to humans, such a method should convey a model's learned classification strategy in a way that is robust to random initializations or spurious correlations in input data.
no code implementations • 31 Oct 2018 • Megha Srivastava, Kalanit Grill-Spector
Because training artificial neural networks from scratch is similar to showing novel objects to humans, we seek to understand the factors influencing the tolerance of CNNs to spatial transformations.