I am Rizal Fathony [full name: Rizal Zaini Ahmad Fathony], a fourth year PhD student at the Department of Computer Science, University of Illinois at Chicago. I work with my advisor Prof. Brian Ziebart in statistical machine learning theory and applications. Before attending PhD program at UIC, I completed Bachelor of Applied Science degree in Statistical Computing from the Institute of Statistics, Jakarta, Indonesia and Master of Science in Computer Science from the University of Illinois at Chicago.
Performance-Aligned Learning Algorithms with Statistical Guarantees
The goal of many prediction tasks in machine learning is to learn a prediction function that minimizes certain loss functions (e.g. zero-one, and ordinal loss) or maximizes certain performance metrics (e.g. accuracy, precision, recall, F-score, and ROC curve). The prevalent techniques in machine learning, probabilistic and large-margin approaches, suffer from the inability to easily incorporate the metrics into the learning process or the lack of statistical guarantee of Fisher consistency. My research focuses on designing learning algorithms that simultaneously align with the learning objective by incorporating the performance metrics or loss functions into the learning process and provide the statistical guarantee of Fisher consistency. My approach in constructing learning algorithms is based on the robust adversarial formulation, i.e. what predictor best maximizes the performance matrix (or minimizes the loss function) in the worst case given the statistical summaries of the empirical distributions. I have designed learning algorithms for many machine learning tasks such as multiclass classification, ordinal regression, weighted bipartite matching, structured prediction, and graphical models.
My future research directions focus on investigating the statistical properties of loss functions, such as stronger statistical guarantees, and the Fisher consistency of structured loss functions, as well as developing learning algorithms for various machine learning tasks, such as structured prediction, graphical models, multi-tasks learning and combining the robust adversarial approach with deep learning.
Some keywords related to my research interest are:
surrogate loss, statistical consistency, robust adversarial formulation, multivariate performance metric, structured prediction, graphical models, game theory, multi-tasks learning, deep learning.
Rizal Fathony, Mohammad Bashiri, Brian D. Ziebart. “Adversarial Surrogate Losses for Ordinal Regression”. Advances in Neural Information Processing Systems 30 (NIPS), 2017. [nips] [pdf] [poster] [code]
Abstract: Ordinal regression seeks class label predictions when the penalty incurred for mistakes increases according to an ordering over the labels. The absolute error is a canonical example. Many existing methods for this task reduce to binary classification problems and employ surrogate losses, such as the hinge loss. We instead derive uniquely defined surrogate ordinal regression loss functions by seeking the predictor that is robust to the worst-case approximations of training data labels, subject to matching certain provided training data statistics. We demonstrate the advantages of our approach over other surrogate losses based on hinge loss approximations using UCI ordinal prediction tasks.
Rizal Fathony, Anqi Liu, Kaiser Asif, Brian D. Ziebart. “Adversarial Multiclass Classification: A Risk Minimization Perspective”. Advances in Neural Information Processing Systems 29 (NIPS), 2016. [nips] [pdf] [poster] [code]
Abstract: Recently proposed adversarial classification methods have shown promising results for cost sensitive and multivariate losses. In contrast with empirical risk minimization (ERM) methods, which use convex surrogate losses to approximate the desired non-convex target loss function, adversarial methods minimize non-convex losses by treating the properties of the training data as being uncertain and worst case within a minimax game. Despite this difference in formulation, we recast adversarial classification under zero-one loss as an ERM method with a novel prescribed loss function. We demonstrate a number of theoretical and practical advantages over the very closely related hinge loss ERM methods. This establishes adversarial classification under the zero-one loss as a method that fills the long standing gap in multiclass hinge loss classification, simultaneously guaranteeing Fisher consistency and universal consistency, while also providing dual parameter sparsity and high accuracy predictions in practice.
Rizal Fathony*, Sima Behpour*, Xinhua Zhang, Brian D. Ziebart. “Efficient and Consistent Adversarial Bipartite Matching”. Submitted to ICML 2018.
Anqi Liu, Rizal Fathony, Brian D. Ziebart. “Kernel Robust Bias-Aware Prediction under Covariate Shift”. ArXiv Preprints, 2016. [arXiv]
NIPS Travel Award 2017
NIPS Travel Award 2016
Won an International Fulbright Master of Science and Technology Scholarship Award (2012 - 2014)
Runner Up Developer at Indonesia Open Source Festival - Android Apps Competition (2010)
Nominee of Research and Development Category at Asia Pacific Information and Communication Technology Award (APICTA), Melbourne, Australia (2009)
Best Research and Development Category at Indonesia Information and Communication Technology Award, Jakarta, Indonesia (2009)
Grantee of a full scholarship and monthly stipend from the Indonesian government during undergraduate study at the Institute of Statistics, Jakarta, Indonesia (2003 - 2007)
Research AssistantDecember 2016 - present
DEPARTMENT OF COMPUTER SCIENCE, UNIVERSITY OF ILLINOIS AT CHICAGO
Research Assistant at Prof. Brian Ziebart’s lab.
Research Intern May 201 - August 2017
TECHNICOLOR RESEARCH, ARTIFICIAL INTELLIGENCE LAB, SAN FRANCISCO BAY AREA
Conducted research in deep learning area especially in Generative Adversarial Networks (GAN):
- Developed a network architecture of GAN for discrete distribution
- Applied Conditional Wasserstein GAN to image colorization tasks
Teaching AssistantAugust 2015 - December 2016
DEPARTMENT OF COMPUTER SCIENCE, UNIVERSITY OF ILLINOIS AT CHICAGO
CS 412 Introduction to Machine Learning (Fall 2016).
CS 412 Introduction to Machine Learning (Spring 2016).
CS 491 Introduction to Machine Learning (Fall 2015).
Statistical Dissemination System Developer January 2008 - June 2012
CENTRAL BUREAU OF STATISTICS INDONESIA
Developed web-based statistical data dissemination and visualization systems for major surveys and censuses conducted by Central Bureau of Statistics Indonesia.
Template Design by W3layouts