Ryan Kiros


PhD Student
Machine Learning Group
Department of Computer Science
University of Toronto

I am a 1st year PhD student under the supervision of Dr. Ruslan Salakhutdinov and Dr. Richard Zemel.
Contact: rkiros [at] cs [dot] toronto [dot] edu

Research

My research interests are in Statistical Machine Learning, Computer Vision and Natural Language Processing.

Education

Publications

Conference Papers

Multimodal Neural Language Models
Ryan Kiros, Ruslan Salakhutdinov, Richard Zemel. ICML, 2014.
[pdf coming soon][code and multimodal word representations coming soon]

Two new neural language models that can be conditioned on other modalities (such as images).
A previous version appeared at the NIPS 2013 Deep Learning Workshop: [pdf]

Training Neural Networks with Stochastic Hessian-Free Optimization
Ryan Kiros. ICLR, 2013.
[pdf] [code] [openreview discussion]

A stochastic version of Martens' Hessian-Free Optimization for training deep nets.

Deep Representations and Codes for Image Auto-Annotation
Ryan Kiros and Csaba Szepesvari. NIPS, 2012.
[pdf] [supplementary] [code and features]

Learning binary codes for image annotation, using a deep unsupervised net.

Regularizers Versus Losses for Nonlinear Dimensionality Reduction
Yaoliang Yu, James Neufeld, Ryan Kiros, Xinhua Zhang, Dale Schuurmans. ICML, 2012.
[pdf] [supplementary]

A unifying view of non-parametric, nonlinear dimensionality reduction + 2 new convex regularizers.

Representation Learning for Sparse, High Dimensional Multi-Label Classification
Ryan Kiros, Axel J. Soto, Evangelos Milios, Vlado Keselj. RSCTC, 2012.
[pdf] [competition page]

5th place (out of 126) in the JRS 2012 data mining competition.

Workshop Papers

In-depth Interactive Visual Exploration for Bridging Unstructured and Structured Document Content
Axel J. Soto, Ryan Kiros, Vlado Keselj, Evangelos Milios. SDM Workshop on Exploratory Data Analysis, 2014.
[pdf coming soon]

Introducing ViTA-SSD, a visual text analytics tool for semi-structured data that utilizes a new deep learning algorithm.

On Linear Embeddings and Unsupervised Feature Learning
Ryan Kiros and Csaba Szepesvari. ICML Representation Learning Workshop, 2012. (Oral)
[pdf] [code]

Combining k-means feature learning with supervised linear t-distributed embeddings.

Technical Reports

Automatic Lung Vessel Segmentation via Stacked Multiscale Feature Learning
Ryan Kiros, Karteek Popuri, Matthew Low, Dana Cobzas, Martin Jagersand. Technical Report, 2013.
[pdf] [code] [competition page] [video] [results viewer]

1st place (out of 24) in the VESSEL12 segmentation challenge.

Master's Thesis

Learning Deep Representations, Embeddings and Codes from the Pixel Level of Natural and Medical Images
Ryan Kiros. University of Alberta, 2013.
[link]

Nominated for MSc Outstanding Thesis Award.

Talks

A Simple Baseline for Segmenting Medical Images. CIFAR Neural Computation and Adaptive Perception Summer School, University of Toronto, 2013.
Deep Representations and Codes for Image Auto-Annotation. University of Guelph, 2012.
On Linear Embeddings and Unsupervised Feature Learning. ICML Representation Learning Workshop, University of Edinburgh, 2012.
Copy Graphs: Compression, Reconstruction and Seed Identification. Canadian Undergraduate Mathematics Conference (CUMC), University of Waterloo, 2010.

Additional Links

Google scholar profile
Occasionally I compete on Kaggle. You can find my profile here.