Zoltán Szabó

Contact Information

Curriculum Vitae [CV (PDF)]

Research Interest

• Dictionary Learning:
  • Group-Structured Dictionary Learning: [39], [37], [35], [34], [29], [28]
    • Online,
    • Arbitrary, overlapping group structures (partition, grid structures, hierarchy,...),
    • Non-convex sparsity-inducing regularization,
    • Handling missing information,
    • +Convex, closed constraints for the dictionary/representation (L₂, L₁, non-negativity, elastic net, fused Lasso, group-norm constraints,...).
  • Independent Subspace Analysis (ISA) and its Extensions (=Independent Process Analysis, IPA: [39], [33], [22]--with a recent ISA review) to:
    • Linear Systems = Integrated Auto-Regressive Moving Average Independent Process Analysis (ARIMA-IPA):
      • Independent Subspace Analysis (ISA, IFSA, MICA, Group ICA, Subspace ICA, IID-IPA): [31],[20], [11], [8],
      • Auto-Regressive Independent Process Analysis (AR-IPA): [25], [12], [7],
      • Blind Subspace Deconvolution (BSSD, MA-IPA): [19], [16], [13],
      • ARMA, ARIMA Models (ARMA-IPA, ARIMA-IPA): [14],
    • Post Nonlinear Models = PNL ISA,...: [17], [15],
    • Complex Valued Models = Complex ARIMA-IPA: [23], [9], [8],
    • Controlled Systems = ARMAX-IPA ('X'=with eXogenous input): [21], [18], [17],
    • Problems with Missing Observations, Partially Observable Systems: [26], [24],
    • Nonparametric (and not Strictly Stationary) Source Dynamics = functional AR-IPA (fAR-IPA): [30], [27].
  • Note:
    • Separation principle can derived for all these models, and
    • The dimension of the source components can be different and unknown ('non-combinatorial' approaches):
      • Basis: ISA Separation Theorem: ISA = ICA + clustering (of the ICA elements) [13],
      • Details for further models: [33], [25], [14], [12], [11],
      • Performance measure: the Amari-index can be extended to the case of different dimensional source components [31], [30], [27], [22].
• Information Theory: multidimensional entropy, mutual information, divergence estimation techniques: [38], [31], [20], [13].
• Kernel Methods:
  • Kernel Regression: [30], [27],
  • 'Kernel Trick': [13], [10], [5].
• Random Projection Techniques, Approximate Nearest Neighbour Preserving Embeddings: [38], [20].
• D-optimal Design: [21], [18], [17].
• Cross-entropy Technique (CE): [20], [7].
• Further applications:
  • collaborative filtering, recommender systems: [34],
  • natural language processing: [37], [35],
  • constrained local models (facial expression estimation, face tracking): [36].

Employment, Education

Publications (.bib); papers on the arXiv (+Atom feed), papers on PASCAL2, papers on MTMT

Hobbies