On the Structures of Representation for the Robustness of Semantic Segmentation to Input Corruption
SS-08:Dynamic Background Reconstruction/Subtraction for Challenging Environments
Visualizing Softmax
Introduction I have been faced with several situations where visualizing the output of a trained classifier has helped explain some interesting behaviors that are present within learned representations. I want to share a simple technique to visualize outputs for SoftMax-based classifiers.
Learning to segment CIFAR10
Import Important Things import torch from torch import nn import torch.nn.functional as F from torch.utils.data import DataLoader from torchvision import transforms from torchvision.models import resnet18 from torchvision.datasets import CIFAR10 from tqdm import tqdm_notebook as tqdm from torchvision.
Speechworks Interview
I was interviewed by Julie Lindsay from Speechworks for a podcast regarding my experiences starting ConvexMind during the CreateX Startup Launch last Summer. There is also an article that summarizes the main points we discussed.
Visualizing Adversarial Attacks
Pixelwise classifiers with Large-margin Gaussian Mixture representation These are several visualizations I dug up from really really messy code, which I won't share just yet. This was some quick and dirty work to test out the feasibility of an idea: I wanted to visualize the effects on the representation space at each pixel during an FGSM attack.
Multi-Basis Input Convolutional Neural Network
I spent some time going through Tensorflow tutorials and was curious what the learned kernels in the convolution layers looked like. After getting a visulization to work for regular RGB input images from CIFAR10, I transformed and created copies of the dataset using other representations: FFT, DCT, and HSV.
