Yumi's Blog

Evaluate uncertainty using ensemble models with likelihood loss and adverserial training

Evaluating the quality of predictive uncertainties is challenging as "ground truth" uncertainty is usually not available. Yet, model's confidence about its estimation is often of interest for researchers. If the model can tell "what it knows" or what is "out of distribution", such infomation gives insights about when the researchers should take the point estimates as their face values.

TensorFlow newbie creates a neural net with a negative log likelihood as a loss

In this blog, I will create a deep learning model that uses the negative log-likelihood of Gaussian distribution as a loss. For this purpose, I will use Tensorflow.

Why not Keras?ΒΆ

Keras has been my first-choice deep learning framework in the last 1 year. However, if you want to create personal loss functions or layers, Keras requires to use backend functions written in either TensorFlow or Theano. As the negative log-likelihood of Gaussian distribution is not one of the available loss in Keras, I need to implement it in Tensorflow which is often my backend. So this motivated me to learn Tensorflow and write everything in Tensorflow rather than mixing up two frameworks.

Learn about Fully Convolutional Networks for semantic segmentation

In this blog post, I will learn a semantic segmentation problem and review fully convolutional networks. In an image for the semantic segmentation, each pixcel is usually labeled with the class of its enclosing object or region. For example, a pixcel might belongs to a road, car, building or a person. The semantic segmentation problem requires to make a classification at every pixel.

Understand Keras's RNN behind the scenes with a sin wave example - Stateful and Stateless prediction -

Recurrent Neural Network (RNN) has been successful in modeling time series data.

People say that RNN is great for modeling sequential data because it is designed to potentially remember the entire history of the time series to predict values. "In theory" this may be true.

But when it comes to implementation of the RNN model in Keras, practitioners need to specify a "length of time series" in batch_shape: