Journey through Fast.AI: II – Columnar data

by Giles Strong

Welcome back to the second part of my journey through the Fast.AI deep-learning course; beginning section here. Last time I gave an example of analysing images, now I’ll move on to working with columnar data.

Columnar data is a form of structured data, meaning that the features of the data are already extracted (in this case into columns), unlike in images or audio where features must be learnt or carefully constructed by hand.

Journey through Fast.AI: I – Introduction and image data

by Giles Strong

For the past few months I’ve been following the Fast.AI Deep-Learning for Coders course. An online series of lectures accompanied with Jupyter notebooks and python library built around PyTorch. The course itself is split into two halves: the first uses a top-down approach to teach state of the art techniques and best practices for deep learning in order to achieve top results on well established problems and datasets, with later lessons delving deeper into the code and mathematics; the second half deals with more with the cutting edge of deep learning, and focuses on less-well-founded problems, such as generative modelling, and recent experimental technologies which are still be developed.

Science in the sun: AMVA4NP’s summer events

by Giles Strong

Summer 2018’s been a busy time for the AMVA4NewPhysics network; we’ve had workshops, outreach events, training sessions, meetings, and many more things. I wanted to go through and pick out a few thinks I was involved in.

Hyper-parameters revisited

by Giles Strong

Introduction

Well folks, it’s been quite a while since my last post; apologies for that, it’s been a busy few months recently.

Towards the end of last year I wrote a post on optimising the hyper parameters (depth, width, learning rate, et cetera) of neural networks. In this post I described how I was trying to use Bayesian methods to ‘quickly’ find useful sets of parameters.

Train-time/test-time data augmentation

by Giles Strong

The week before last I was presenting an update of some of my analysis work to the rest of my group. The work involved developing a neural-network to classify particle-collisions at the LHC.

Higgs Hacking

by Giles Strong

A few days before I returned from CERN at the beginning of the month, I attended a talk on the upcoming TrackML challenge. This is a competition beginning this month in which members of the public will be invited to try and find a solution to the quite tricky problem of accurate reconstruction of particle trajectories in the collisions at the LHC. The various detectors simply record the hits where particles pass by, however to make use of this data, the hits in surrounding detector layers must be combined into a single flight path, called a track.

Staying at CERN

by Giles Strong

Bonjour! As I write, I’m three weeks into my month long secondment at CERN, near Geneva. CERN, home to the Large Hadron Collider, is the world’s largest particle-physics research centre. It is also the location of the CMS experiment, which I work on.

Efficiency revisited

by Giles Strong

Cover photo unrelated – it’s just some rad fractal broccoli.

Just over one and a half years ago I wrote a post on some of the tips and tricks I’d found useful in trying to organise myself and improve my efficiency. Searching for a post topic, it was suggested that I revisit this to compare how my workload and approaches have changed, so here goes!

Adjusting hyper-parameters: First step into Bayesian optimisation of DNNs

by Giles Strong

A few months ago I wrote about some work I was doing on improving the way a certain kind of particle is detected at CMS, by replacing the existing algorithm with a neural network. I recently resumed this work and have now got to the point where I show significant improvement over the existing method. The design of the neural network, however, was one that I imported from some other work, and what I want to do is to adjust it to better suit my problem.