Machine Learning with Unbalanced Data
In many practical applications, datasets used for training neural networks
are highly unbalanced or scarce in volume. Examples can be
found in medical imaging, where data from common diseases are significanlty more than that of rare diseases, or in financial transactions with much more healthy transactions than fradulent ones, or in the
analysis of network traces with hackers traffic constituting only a small amount of communications. Training deep neural networks with unbalanced data tends to bias the network towards the classes that contain more data.
In our research, we address these shortcomings by data augmentation:
Complexity reduction of Deep Neural Networks
Scarcity of data is a challenge in training large neural networks.
Training large networks with small data might result in overfitting. Drop-out has been proposed to address this problem by removing some of the neurons during the training phase.
In our research on drop-out, we represent the neural net with a graphical
model and minimize the Ising energy to identify the neurons that should be dropped during the training and also during inference.
In our approach, neurons that operate in the saturated region of the sigmoid function are prone to be dropped.
- Reduction of network complexity using drop-out
- Salehinejad H, Valaee S. "ISING-DROPOUT: A Regularization Method for Training and Compression of Deep Neural Networks," IEEE International Conference on Acoustics, Speech and Signal Processing (IEEE ICASSP), 2019, UK.
Dropping Saturated Neurons
Localization of Wireless Terminals
We have developed an accurate location estimation technique for indoor environment using compressive sensing. The proposed algorithm has been developed on HP PDA and Android devices and tested in Bahen Centre at the University of Toronto, Bayview Village shopping mall in North Toronto, and the Canadian National Institute for Blind (CNIB). Currently, we are using machine learning and channel state information for location estimation.
See our demo
[very large file] and our patents:
- S. Valaee, C. Feng, and A. W. S. Au, "System, Method, and Computer Program for Anonymous Localization," US 14/576,586.
- S. Valaee and C. Feng, "System, method and computer program for dynamic generation of a radio map for indoor positioning of mobile devices," US 13/927,510.
- S. Shahidi and S. Valaee, "Indoor localization using crowdsourced data," US 14/745,873
- Some of our recent publications:
- Muhammed T. Rahman, Navid Tadayon, Shuo Han, and Shahrokh Valaee, "LocHunt: Angle of Arrival Based Location Estimation in Harsh Multipath Environments", in IEEE Globecom 2018.
- Muhammed Tahsin Rahman, Shuo Han, Navid Tadayon, and Shahrokh Valaee,
"Ising Model Formulation of Outlier Rejection, with Applications in WiFi based Positioning", in IEEE ICASSP 2019.
- Navid Tadayon, Muhammed T. Rahman, Shuo Han, Shahrokh Valaee, Wei Yu,
"Decimeter Ranging with Channel State Information".
Link to Arxiv
Our research in vehicular networks focuses on interference management in wireless channel. The research includes both 802.11p Wireless Access in Vehicular Environment (WAVE), and C-V2X. We have shown that 802.11p suffers from significant interference that limits the number of vehicles that can participate in the network. Using the concept of Pseudo Orthogonal Codes (POC), we have designed effective distributed schedulers that are applicable to both 802.11p and C-V2X. Currently, we are working on scheduling sidelinks in device-to-device (D2D) and LTE-V2X technology. We have designed effective
resource management schemes for multichannel operation of V2X and D2D.
- Some of our recent publications:
- Zahra Naghsh, Mohammad Javad-Kalbasi, and Shahrokh Valaee,
" Digitally Annealed Solution for the Maximum Clique Problem with Critical Application in Cellular V2X ",
in IEEE ICC 2019.
- Zahra Naghsh and Shahrokh Valaee, " Delay-aware Conflict-free Scheduling for LTE-V, Sidelink 5G V2X Vehicular Communication, in Highways ",
The 52nd Asilomar Conference on Signals, Systems and Computers, 2018.
- Zahra Naghsh and Shahrokh Valaee,
" MUCS: A New Multichannel Conflict-Free Link Scheduler for Cellular V2X Systems", in IEEE ICC 2018 Wireless Networking Symposium.
- Here is a demo of an earlier version of our algorithms on
Contiguous Scheduling in C-V2X