I am a PhD candidate at MILA advised by Yoshua Bengio and co-advised by Victor Lempitsky of Skoltech. My scientific interests span computer vision and machine learning. Nowadays I’m mostly working on artificial neural networks.
If you really want to, you can access my typed curriculum vitae, although I tend to update the LinkedIn profile more frequently.
Here is how you would contact me:
DeepWarp: Photorealistic Image Resynthesis for Gaze Manipulation
Yaroslav Ganin, Daniil Kononenko, Diana Sungatullina, Victor Lempitsky
A deep learning system for real-time gaze redirection. We propose a novel architecture that predicts a warping transformation (much like remap in OpenCV) and thus does not suffer from problems (information loss, regression-to-mean, etc.) of conventional encoder-decoder approaches.
Accepted to ECCV’16.
Unsupervised Domain Adaptation by Backpropagation
Yaroslav Ganin, Victor Lempitsky
Deep neural networks can be adapted for the new domains by incorporating unlabeled data into the learning procedure. Very easy to implement, yet gives state-of-the-art results on the Office dataset and several other benchmarks.
Links to the datasets: MNIST-M, SynNumbers. I’m also providing an unpacked version of MNIST-M (courtesy of Konstantinos Bousmalis).
Accepted to ICML’15.
Speeding-up Convolutional Neural Networks Using Fine-tuned CP-Decomposition
Vadim Lebedev, Yaroslav Ganin, Maksim Rakhuba, Ivan Oseledets, Victor Lempitsky
We propose a simple two-step approach for speeding up convolution layers within large convolutional neural networks based on tensor decomposition and discriminative fine- tuning.
Accepted to ICRL’15.
$ N^4 $-Fields: Neural Network Nearest Neighbor Fields for Image Transforms
Yaroslav Ganin, Victor Lempitsky
__N__eural __N__etworks and __N__earest __N__eighbours search (hence $ N^4 $) fused together to form a general approach for solving such tasks as edge detection and segmentation (probably some others too).
Accepted for an oral presentation at ACCV’14 (4% acceptance rate).
Efficient Segmentation Trees on the GPU
Yaroslav Ganin
Here’s how one may get a segmentation tree via Borůvka’s MST algorithm. No wheels were reinvented in the making of this project. Pure Thrust (well, almost).
Accepted for a poster presentation at GTC’12.
You can find the code in the CUDA Toolkit bundle.
Here are some older projects that I was working on during my studies at MSU:
I was a TA for the following classes:
Under construction.