Research

RESEARCH TOPICS

My research lies at the intersection of optimization, learning, signal processing, and wireless communications.

Interpretable Models in Machine Learning

Models at the heart of machine learning and artificial intelligence. ML follows the black-box model (i.e., no interpretability). Interpreabilty is achieved by designing model which facilitates finding causal relations. We instantiate these relations by finding sufficient conditions which bind (in the sense of implications in mathematical logic) outcomes of specific random variables and extract valuable relations from the data. We also propose an efficient algorithm for computing the model.

We also investigate applications of this framework in: recommendation systems, bioinformatics (DNA analysis), and beamforming for large mmWave MIMO

SELECTED RELEVANT WORK:

  • H. Ghauch, H. Shokri, M. Skoglund, C. Fischione, A. Sayed, "Learning Kolmogorov Models for Binary Random Variables", ICML 2018
  • H. Ghauch, M. Skoglund, C. Fischione, A. Sayed, "Learning elementary representations of random variables", Journal of Machine Learning Research (in preparation)

Learning-based Resource Allocation

Many wireless communication systems are highly correlated: millimeter-wave communication, line-of-sight channels. We leverage this inherent property using machine learning to perform resource allocation, based on position information. This drastically reduces the overhead needed estimating the large number of channels in the network

SELECTED RELEVANT WORK:

Optimization for mmWave Communication

mmWave bands have ~200 times more spectrum than traditional cellular systems. Hundreds of antennas are needed at the transmitter and receiver (to combat the sever pathloss of mmWave channels), making the power consumption too high. Instead, a Hybrid Analog-Digital Architecture was proposed instead (overview slide)

Our works address several fundamental problems such as efficient channels estimation, algorithmic solution to the hybrid precoding problem, and channel tracking in case of mobility

SELECTED RELEVANT WORK (picking up on google scholar):

Low-overhead Coordination in 5G Cellular Networks

5G cellular systems require ever increasing data rates. Distributed coordination among base station increases network throughput. However, the coordination overhead is a limiting factor.

Our works proposed several classes of distributed optimization with fast convergence, to reduce the coordination overhead (overview slide)

SELECTED RELEVANT WORK:

Optimizing Cloud Radio-Access Networks

Most of the gains in data rates are from densification (more antennas/base stations). Cloud RAN allows for tight coordination of antennas, in a cheap manner (because it is centralized).

Our works investigate several optimization techniques to address problems such as beamforming design, and user assignment (overview slide)

SELECTED RELEVANT WORK: