Demo: Kernel methods for machine learning applications

[raamana]

Title
Kernel methods library for machine learning applications

Presentor and Affiliation
Pradeep Reddy Raamana,
Rotman Research Institute, Baycrest Health Sciences,
University of Toronto, Toronto, Canada.

Collaborators
TBD

Github Link (if applicable)
kernelmethods

Abstract (max. 200 words):
kernelmethods is a pure python library defining modular classes that provides basic kernel methods and an intuitive interface for advanced functionality such as composite and hyper kernels. This library fills an important void in the ever-growing python-based machine learning ecosystem, where users are limited to few predefined kernels without the ability to customize or extend them for their own applications. This library defines the KernelMatrix class that is central to all the kernel methods. As it is a key bridge between input data and kernel learning algorithms, it is designed to be highly usable and extensible to different applications and data types. Kernel operations implemented are normalization, centering, product, alignment, linear combination and ranking. Convenience classes, such as Kernel{Set,Bucket}, are designed for easy management of a large collection of kernels. Dealing with diverse kernels and their fusion is necessary for automatic kernel selection in applications such as Multiple Kernel Learning. Besides numerical kernels, we designed this library to provide categorical, string and graph kernels, with the same attractive properties of intuitive and highly-testable API. Besides non-numerical kernels, we aim to provide a deeply extensible framework for arbitrary input data types, such as sequences and trees, via pyradigm. Moreover, drop-in Estimator classes are provided for seamless usage in scikit-learn ecosystem.

Preferred Session
Demo: New advances in open neuroimaging methods

Additional Context
With the ever-increasing diversity of datasets and diseases in neuroimaging research, having the ability to learn the kernel on one's own dataset/application (instead of ramming a pre-defined kernel that users can't inspect or customize) is crucial to optimize the performance of a kernel-based machine learning technique such as SVM, MKL and the like.

[raamana]

Hi @TimVanMourik, how would I know if I am selected to present or not? It helps me to prepare if you give me a 7-10 days notice. thanks.

[emdupre]

Hi @raamana ! Tim is finalizing the schedule this week -- all talks will be announced on the 24th (this Friday) as outlined in the talk application.

Hope that helps, and thanks for submitting ! Looks like really interesting work 🌺

[raamana]

My bad, I missed it. Friday would be great. Thanks.

Thanks, this library is something I am quite proud of building. Hoping to improve it with other kernel methods enthusiasts, esp. for graph type data which neuroscientists are familiar with :).

[raamana]

I will have the demo notebooks uploaded early next week.

Roughly, this can be thought of as the pure-python equivalent for jKernelMachines in Java and kernlab in R

[TimVanMourik]

Hi @raamana, I’m happy to tell you that we’d like to host your presentation as a lightning talk in the OSR in the Machine learning in Neuroscience session. This will be a talk of 5 minutes + 5 minutes of questions. We decided to rebrand one session of lightning talks to a machine learning theme as a result of many applications around this theme. We cannot give you a slot in your preferred session due to the very high number of applications.

We’ll update the program in the ReadMe.md shortly. We’d much appreciate it if you could submit slides and other presentation material to the presentations folder by means of a Pull Request to this repository, preferably but not necessarily before the presentation.

[raamana]

Thanks Tim - appreciate it. I understand you all had a tough job sorting this all out.

Looking fwd to presenting it.