Welcome to my blog. I use it to document what I learn in the rapidly changing field of AI, mostly through the lens of multimodality: speech language models, text-to-speech, speech recognition and large language models.
I build AI products at GetVocal AI in Paris. More about me.
Accuracy is one of the most used metrics to evaluate classification tasks in machine learning. It is the ratio of the number of correct predictions to the total number of examples. It is simple to understand and compute, which makes it an easy evaluation metric to optimize but it has...
Over the past few months, I have come across a plethora of questions related to dataset creation for speech projects. I could not find a concise resource detailing all the necessary factors that need to be put in place to have a well balanced, unbiased and clean speech corpus. Many...
How the forward-backward algorithm computes the Connectionist Temporal Classification loss, step by step.
Attention-based networks have been shown to outperform recurrent neural networks and its variants for various deep learning tasks including Machine Translation, Speech, and even Visio-Linguistic tasks. The Transformer is a model, at the forefront of using only self-attention in its architecture, avoiding recurrence and enabling parallel computations.
A Neural network is a universal function approximator, so in theory it is possible to learn any function using a neural network. As K-nearest neighbor is a method of predicting the label of a new datapoint from the test set, it is possible to express its prediction function as a...
Neural networks have a high tendency to overfit on training data, especially when the examples are few and the network has a large capacity. There is even a famous quote which says, A popular deep learning adage is that “If your neural network is not overfitting, then it is not...
Softmax function is one of the major functions used in classification models. It is usually introduced early in a machine learning class. It takes as input a real-valued vector of length, d and normalizes it into a probability distribution. It is easy to understand and interpret but at its core...
This article was inspired by Angela Fan’s Talk on Efficient Transformers at AMMI Deep NLP class, 2020. Angela Fan is a PhD candidate with Facebook AI in France.