Liked my research? Have a question? Contact Me!

Data scicence

Using machine learning to make skillful predictions of the winter time North Atlantic Oscillation
(Papers submitted to ASL)
Humpback whale identification
(https://www.kaggle.com/c/humpback-whale-identification)

In this competition, we need to build an algorithm to identify whale species in images over 25,000 images. This competition aims to aid whale conservation efforts as scientists often use photo surveillance systems to monitor ocean activity. These photos often capture the tail of the whales and it is useful for scientists to be able to quickly and accurately identify the exact whale species from a photo of the tail.

In this competition, I used a resnet style convolution neural network to train (using SGDR with lr annealing) and identify the whale species. At the time of writing, I have achieved a mean average precision of 0.411. The accuracy can potentially be further improved by implementing a model with warm restarts, using bounding boxes and further data augmentations.

For my kaggle, see https://www.kaggle.com/scottyiu

MNIST digit recognition
(https://www.kaggle.com/c/digit-recognizer)

In this competition, we need to build an algorithm to identify human written digits from the MNIST dataset. This is meant as a playground to test new algorithms.

In this competition, I used a resnet style convolution neural network to train (using SGDR with lr annealing) and identify the digits. At the time of writing, I have achieved a accuracy of over 99.5%. The accuracy can potentially be further improved by implementing a model with warm restarts and longer training.

For my kaggle, see https://www.kaggle.com/scottyiu

Fast.ai courses
(https://www.fast.ai/)

Fast.ai is a deep learning package built on top of pytorch and is capable of implementing neural networks quickly and easily. They run a MOOC on data science and deep learning.

I have finished the deep learning course 1 which covers using fast.ai for image recognition, NLP, structured data and collaborative filtering.

Cambridge Spark advanced machine learning courses
(https://cambridgespark.com/)

Cambridge spark provides an advanced machine learning course that demonstrates the power of using Keras (with theano backend) in creating neural networks to tackle data science problems.

These courses were held in London.