Professor Ghodsi's research interests lie at the interface of statistics and computer science. They span a variety of areas in computational statistics particularly in the areas of machine learning and probabilistic modelling.
He studies theoretical frameworks and develops new machine-learning algorithms for analyzing large-scale data sets, with applications in data mining, pattern recognition, robotics, computer vision, sequential decision making, and bioinformatics.
Some of Professor Ghodsi's current work focuses on the dimensionality reduction (manifold learning) problem. Dimensionality reduction addresses the problem of dealing with complex data by mapping high-dimensional data into fewer dimensions. Many problems of scientific interest that require the analysis of very large high-dimensional data sets can benefit from dimensionality reduction techniques.
An essential part of the information that is totally ignored by existing dimensionality reduction techniques is knowledge about the sequence of the observations and the actions between data points. Professor Ghodsi recently co-developed a new dimensionality reduction technique called Action Respecting Embedding (ARE) which exploits this additional information, successfully translating actions into meaningful and interpretable low-dimensional representations. This led to novel solutions to sequential decision problems such as planning (i.e., finding a sequence of actions to achieve a particular outcome) and localization (i.e., maintaining a representation of one's location). Unlike existing techniques, this approach requires no expert knowledge about the domain to find effective solutions. Professor Ghodsi is working to refine this new technique and to make it more efficient and scalable. This is a potential solution to a large number of problems of scientific interest, including industrial processes and inventory management.
On the more theoretical side, Professor Ghodsi is exploring and formalizing nonlinear dimensionality reduction techniques as probabilistic models. He is addressing the problem of how such models should be constructed, and how they should respond when data is missing. This has many potential uses in fields such as physics, economics, and medicine, where meaningful information must be extracted from large data sets.
Professor Ghodsi is currently a member of the Centre for Computational Mathematics in Industry and Commerce, and the Artificial Intelligence Research Group at the University of Waterloo. He has worked in two other world-class research environments at the University of Toronto and the University of Alberta.
In particular, over the past three years he has spent a significant amount of research time at the Probabilistic and Statistical Inference Group at the University of Toronto and at the Alberta Ingenuity Centre for Machine Learning at the University of Alberta, where he collaborated on statistical machine-learning methods applied to robotics and pattern recognition problems. Since 1992 he has spent five years in industry where he was involved with both software design and implementation.
- Babak Alipanahi Ramandi, Nathan Krislock, Ali Ghodsi, Henry Wolkowicz, Logan Donaldson and Ming Li, Determining Protein Structures from NOE Distance Constraints by Semidefinite Programming, to appear in The Journal of Computational Biology 2012
- Ahmed Farahat, Ali Ghodsi, and Mohamed Kamel. An Efficient Greedy Method for Unsupervised Feature Selection, In proceedings of the Eleventh IEEE International Conference on Data Mining, 2011 (ICDM 2011).
- Ahmed K. Farehat, Ali Ghodsi, and Mohamed S. Kamel. A novel greedy algorithm for Nystrom approximation, Fourteenth International Conference on Artificial Intelligence and Statistics (AI & Statistics 2011)
- Elnaz Barshan, Ali Ghodsi, Zohreh Azimifar, and Mansoor Zolghadri. Supervised Principal Component Analysis: Visualization, Classification and Regression on Subspaces and and Regression on Subspaces and Submanifolds , Journal of Pattern Recognition.
- Stefan Pintilie, and Ali Ghodsi. Conformal Mapping by Computationally Efficient Methods, In proceedings of Twenty-Fourth National Conference on Artificial Intelligence (AAAI 2010).
- Michael Biggs, Ali Ghodsi, and Stephen Vavasis. Nonnegative Matrix Factorization via Rank-One Downdate, In proceedings of The 25th International Conference on Machine Learning (ICML 2008).
- Babak Alipanahi, Michael Biggs, and Ali Ghodsi. Distance Metric Learning Vs. Fisher Discriminant Analysis, In proceedings of Twenty-Third National Conference on Artificial Intelligence (AAAI 2008).
- Mu Zhu and Ali Ghodsi. Automatic Dimensionality Selection from the Scree Plot via the Use of Profile Likelihood, Journal of Computational Statistics and Data Analysis 2005.
- Ali Ghodsi, Jiayuan Huang, Finnegan Southey, and Dale Schuurmans. Tangent-Corrected Embedding, In proceedings of International Conference on Computer Vision and Pattern Recognition (CVPR 2005).
- Michael Bowling, Ali Ghodsi, and Dana Wilkinson. Action Respecting Embedding, In proceedings of The 22nd International Conference on Machine Learning (ICML 2005).