Scaleable input gradient regularization for adversarial robustness
Finlay, Chris, and Adam M. Oberman. "Scaleable input gradient regularization for adversarial robustness." Machine Learning with Applications 3 (2021): 100017.
CV
Education
- Ph.D in Mathematics, McGill University, 2019
- B.Sc. (Hons) in Applied Mathematics, University of Alberta, 2011
- Dip. Music in Jazz Performance, Grant MacEwan, 2007
Work experience
- Post-doctoral researcher, McGill University, Jan 2020 - June 2020
- Visiting research fellow, Institute for Pure & Applied Mathematics, UCLA, Los Angeles, California, March 2020 - June 2020
- Data analyst, Canadian Forest Service, Natural Resources Canada, Oct 2012 - Aug 2014
- developed statistical models to predict wood quality from historical climate data
- created a C++ library to implement these models for use in forest modeling software
- constructed, cleaned and maintained a massive database of historical tree-ring data
- modelled the pine beetle invasion using techniques from applied math
- synthesized climate model forecasts with ensemble modeling methods (in R)
- Guitar instructor, Long & McQuade, Sept 2010 - June 2013
- taught private music lessons to children, teens, and adults, tailored to student ability
- Web designer (short-term contract), Aerosol Research Lab of Alberta, Fall 2012
- design an online web application (HTML/Javascript) implementing aerosol deposition models in the human lung, to help physicians find proper aerosol dosages (for drug delivery by inhaler)
- Patent analyst, Lambert Intellectual Property Law, Summer 2012
- research patent viability, assist drafting & filing patents
- Student researcher, Dept of Mathematics, University of Alberta, Summers 2009 & 2010
- implement mathematical models describing tumour growth (in MATLAB)
- supervised by Thomas Hillen
Skills
- Strong coding and development background. Expert in Python, C++, MATLAB and R
- Expert in developing deep learning architectures for computer vision: strong PyTorch background and comfortable in TensorFlow
- Long-suffering Linux user, with over 10 years experience developing in CLI and remote environments
- Strong background in optimization algorithms
- Experience developing and maintaing a codebase for a team of researchers and students
- for example, see my template for computer vision experiments, which is heavily used by members of the Oberman lab
Publications
How to train your Neural ODE: the world of Jacobian and kinetic regularization
Finlay, Chris, Jörn-Henrik Jacobsen, Levon Nurbekyan, and Adam M. Oberman. "How to train your Neural ODE: the world of Jacobian and kinetic regularization." International Conference on Machine Learning, 2020..
A principled approach for generating adversarial images under non-smooth dissimilarity metrics
Aram-Alexandre Pooladian, Chris Finlay, Tim Hoheisel, and Adam M. Oberman. "A principled approach for generating adversarial images under non-smooth dissimilarity metrics." 23rd International Conference on Artificial Intelligence and Statistics (2020).
The LogBarrier adversarial attack: making effective use of decision boundary information
Finlay, Chris, Aram-Alexandre Pooladian and Adam M. Oberman. "The LogBarrier adversarial attack: making effective use of decision boundary information." International Conference on Computer Vision, 2019.
Improved accuracy of monotone finite difference schemes on point clouds and regular grids
Finlay, Chris, and Adam M. Oberman. "Improved accuracy of monotone finite difference schemes on point clouds and regular grids." SIAM Journal on Scientific Computing 49, no. 15 (2019): 3097-3117.
Approximate homogenization of convex nonlinear elliptic PDEs
Finlay, Chris, and Adam M. Oberman. "Approximate homogenization of convex nonlinear elliptic PDEs." Communications in Mathematical Sciences 16, no. 7 (2018): 1895-1906.
Approximate homogenization of fully nonlinear elliptic PDEs: estimates and results for Pucci type equations
Finlay, Chris, and Adam M. Oberman. "Approximate homogenization of fully nonlinear elliptic PDEs: estimates and results for Pucci type equations." Journal of Scientific Computing 77, no. 2 (2018): 936-949.
Annual ring density for lodgepole pine as derived from models for earlywood density, latewood density and latewood proportion
Sattler, Derek F., Chris Finlay, and James D. Stewart. "Annual ring density for lodgepole pine as derived from models for earlywood density, latewood density and latewood proportion." Forestry: An International Journal of Forest Research 88, no. 5 (2015): 622-632.
Are more complicated tumour control probability models better?
Gong, Jiafen, Mairon M. Dos Santos, Chris Finlay, and Thomas Hillen. "Are more complicated tumour control probability models better?" Mathematical Medicine and Biology: a journal of the IMA 30, no. 1 (2011): 1-19.
From cell population models to tumor control probability: including cell cycle effects
Hillen, Thomas, Gerda De Vries, Jiafen Gong, and Chris Finlay. "From cell population models to tumor control probability: including cell cycle effects." Acta Oncologica 49, no. 8 (2010): 1315-1323.
Talks
Training neural ODEs for density estimation
Talk at IPAM: Workshop on PDEs and Inverse Problems in Machine Learning, LA, California
Regularized neural ODEs for density estimation with normalizing flows
Talk at MtlOPT working group, Montreal, Quebec
The LogBarrier adversarial attack
Poster at International Conference on Computer Vision (ICCV), Seoul, Korea
On some applied problems using nonlinear elliptic PDEs
Talk at McGill University, Montreal, Quebec
Scaleable input regularization for adversarial robustness
Poster at DLRL Summer School, University of Alberta, Edmonton, Alberta
TV Regularization leads to adversarially robust and accurate neural networks
Talk at Level Set Collective, UCLA, LA, California
Teaching
Service and leadership
- President of the Graduate Student Association of Mathematics and Statistics, McGill University, 2015-2017
Not a robot
- I’m a very active birder [for other twitchers, here’s my eBird profile]
- These days I play a lot of pedal steel guitar, mostly with Night Jar
- I sing tenor in the Monday Night Choir, and act as secretary for that organization