SLIP: Learning to Predict in Unknown Dynamical Systems with Long-Term Memory

We present an efficient and practical (polynomial time) algorithm for online prediction in unknown and partially observed linear dynamical systems (LDS) under stochastic noise. When the system parameters are known, the optimal linear predictor is the Kalman filter. However, the performance of existing predictive models is poor in important classes of LDS that are only marginally stable and exhibit long-term forecast memory. We tackle this problem through bounding the generalized Kolmogorov width of the Kalman filter model by spectral methods and conducting tight convex relaxation. We provide a finite-sample analysis, showing that our algorithm competes with Kalman filter in hindsight with only logarithmic regret. Our regret analysis relies on Mendelson's small-ball method, providing sharp error bounds without concentration, boundedness, or exponential forgetting assumptions. We also give experimental results demonstrating that our algorithm outperforms state-of-the-art methods. Our theoretical and experimental results shed light on the conditions required for efficient probably approximately correct (PAC) learning of the Kalman filter from partially observed data.

[1]  Martin J. Wainwright,et al.  From Gauss to Kolmogorov: Localized Measures of Complexity for Ellipses , 2018, Electronic Journal of Statistics.

[2]  Ambuj Tewari,et al.  Optimism-Based Adaptive Regulation of Linear-Quadratic Systems , 2017, IEEE Transactions on Automatic Control.

[3]  Bin Yu RATES OF CONVERGENCE FOR EMPIRICAL PROCESSES OF STATIONARY MIXING SEQUENCES , 1994 .

[4]  Daphne Koller,et al.  Continuous Time Bayesian Networks , 2012, UAI.

[5]  D. Donoho,et al.  Minimax Risk Over Hyperrectangles, and Implications , 1990 .

[6]  A. Zeevi,et al.  NONASYMPTOTIC BOUNDS FOR AUTOREGRESSIVE TIME SERIES MODELING , 2001 .

[7]  Ambuj Tewari,et al.  Finite Time Identification in Unstable Linear Systems , 2017, Autom..

[8]  T. Lai,et al.  Recursive identification and adaptive prediction in linear stochastic systems , 1991 .

[9]  Hugh F. Durrant-Whyte,et al.  Simultaneous localization and mapping: part I , 2006, IEEE Robotics & Automation Magazine.

[10]  Shahar Mendelson,et al.  Learning without Concentration , 2014, COLT.

[11]  Stuart J. Russell,et al.  Why are DBNs sparse? , 2010, AISTATS.

[12]  Michael I. Jordan,et al.  Learning Without Mixing: Towards A Sharp Analysis of Linear System Identification , 2018, COLT.

[13]  Wayne A. Fuller,et al.  Predictors for the first-order autoregressive process , 1980 .

[14]  George J. Pappas,et al.  Finite Sample Analysis of Stochastic System Identification , 2019, 2019 IEEE 58th Conference on Decision and Control (CDC).

[15]  Yi Zhang,et al.  Spectral Filtering for General Linear Dynamical Systems , 2018, NeurIPS.

[16]  Sham M. Kakade,et al.  A Linear Dynamical System Model for Text , 2015, ICML.

[17]  Nikolai Matni,et al.  Regret Bounds for Robust Adaptive Control of the Linear Quadratic Regulator , 2018, NeurIPS.

[18]  Mehryar Mohri,et al.  Rademacher Complexity Bounds for Non-I.I.D. Processes , 2008, NIPS.

[19]  Adel Javanmard,et al.  The minimax risk of truncated series estimators for symmetric convex polytopes , 2012, 2012 IEEE International Symposium on Information Theory Proceedings.

[20]  Nassir Navab,et al.  Long Short-Term Memory Kalman Filters: Recurrent Neural Estimators for Pose Regularization , 2017, 2017 IEEE International Conference on Computer Vision (ICCV).

[21]  L. Ljung Convergence of an Adaptive Filter Algorithm , 1978 .

[22]  G. Lorentz,et al.  Constructive approximation : advanced problems , 1996 .

[23]  Fuzhen Zhang The Schur complement and its applications , 2005 .

[24]  Elad Hazan,et al.  Introduction to Online Convex Optimization , 2016, Found. Trends Optim..

[25]  A. Pinkus n-Widths in Approximation Theory , 1985 .

[26]  Michel Verhaegen,et al.  Identification of structured state-space models , 2018, Autom..

[27]  O. Papaspiliopoulos High-Dimensional Probability: An Introduction with Applications in Data Science , 2020 .

[28]  I. Ibragimov,et al.  On density estimation in the view of Kolmogorov's ideas in approximation theory , 1990 .

[29]  Alessandro Lazaric,et al.  Improved Regret Bounds for Thompson Sampling in Linear Quadratic Control Problems , 2018, ICML.

[30]  Tengyu Ma,et al.  Gradient Descent Learns Linear Dynamical Systems , 2016, J. Mach. Learn. Res..

[31]  Haipeng Luo,et al.  Logistic Regression: The Importance of Being Improper , 2018, COLT.

[32]  George J. Pappas,et al.  Online Learning of the Kalman Filter With Logarithmic Regret , 2020, IEEE Transactions on Automatic Control.

[33]  Alexander Rakhlin,et al.  Near optimal finite time identification of arbitrary linear dynamical systems , 2018, ICML.

[34]  C. Z. Wei Adaptive Prediction by Least Squares Predictors in Stochastic Regression Models with Applications to Time Series , 1987 .

[35]  Csaba Szepesvári,et al.  Improved Algorithms for Linear Stochastic Bandits , 2011, NIPS.

[36]  Shie Mannor,et al.  On-Line Learning of Linear Dynamical Systems: Exponential Forgetting in Kalman Filters , 2018, AAAI.

[37]  Sham M. Kakade,et al.  A tail inequality for quadratic forms of subgaussian random vectors , 2011, ArXiv.

[38]  Alex Townsend,et al.  On the Singular Values of Matrices with Displacement Structure , 2016, SIAM J. Matrix Anal. Appl..

[39]  Karan Singh,et al.  Learning Linear Dynamical Systems via Spectral Filtering , 2017, NIPS.

[40]  Yi Ouyang,et al.  Learning-based Control of Unknown Linear Systems with Thompson Sampling , 2017, ArXiv.

[41]  S. Y. Chen,et al.  Kalman Filter for Robot Vision: A Survey , 2012, IEEE Transactions on Industrial Electronics.

[42]  Roman Vershynin,et al.  High-Dimensional Probability , 2018 .

[43]  Max Simchowitz,et al.  Naive Exploration is Optimal for Online LQR , 2020, ICML.

[44]  W. Fuller,et al.  Properties of Predictors for Autoregressive Time Series , 1981 .

[45]  Benjamin Recht,et al.  Certainty Equivalent Control of LQR is Efficient , 2019, ArXiv.

[46]  Stuart J. Russell,et al.  Probabilistic detection of short events, with application to critical care monitoring , 2008, NIPS.

[47]  Mehryar Mohri,et al.  Generalization bounds for non-stationary mixing processes , 2016, Machine Learning.

[48]  Karan Singh,et al.  No-Regret Prediction in Marginally Stable Systems , 2020, COLT.

[49]  Yihong Wu,et al.  Volume ratio, sparsity, and minimaxity under unitarily invariant norms , 2013, 2013 IEEE International Symposium on Information Theory.

[50]  S. Frick,et al.  Compressed Sensing , 2014, Computer Vision, A Reference Guide.

[51]  T. Lai,et al.  Least Squares Estimates in Stochastic Regression Models with Applications to Identification and Control of Dynamic Systems , 1982 .

[52]  V. Kozyakin On accuracy of approximation of the spectral radius by the Gelfand formula , 2008, 0810.2856.

[53]  Martin J. Wainwright,et al.  The Local Geometry of Testing in Ellipses: Tight Control via Localized Kolmogorov Widths , 2017, IEEE Transactions on Information Theory.

[54]  R.S. Parker,et al.  A model-based algorithm for blood glucose control in Type I diabetic patients , 1999, IEEE Transactions on Biomedical Engineering.

[55]  Max Simchowitz,et al.  Learning Linear Dynamical Systems with Semi-Parametric Least Squares , 2019, COLT.