Roni Khardon Professor Department of Computer Science Indiana University, Bloomington Research Interests and Projects: My interests are in the areas of Machine Learning and Data Mining, Artificial Intelligence, and Efficient Algorithms and my research explores theoretical questions, empirical questions, and applications. Most recently my work has focused on problems in two broad areas: Theory, Algorithms and Applications of Graphical models and Stochastic Planning and Reinforcement Learning and their Relation to Probabilistic Inference.

Software and Data

• DiSProD (Palash Chatterjee with Ashutosh Chapagain) Code for planning/control in continuous and hybrid spaces is provided through Palash Chatterjee's github site. For context and description please see:
  DiSProD: Differentiable Symbolic Propagation of Distributions for Planning , IJCAI 2023.

  Download code through:  https://github.com/pecey/DiSProD

• DLM and Variational Methods for BNN (Yadi Wei) Code for learning and Bayesian Neural Networks and evaluating the quality of uncertainty quantofication they provide, including direct loss minimization and variational methods, is provided through Yadi Wei's github site. For context and description please see:
  On the Performance of Direct Loss Minimization for Bayesian Neural Networks , ICBINB Workshop, 2022.

  Download code through: https://github.com/weiyadi/dlm_bnn

• Approximate Inference for Stochastic Planning (Zhennan Wu) Code for stochastic planning through algorithms for problebilistic inference (including mean field variational inference, belief propagation, SOGBOFA, and a new method collapsted state variational inference) is provided through Zhennan Wu's github site. For context and description please see:
  Approximate Inference for Stochastic Planning in Factored Space , PGM, 2022.

  Download code through:  https://github.com/Zhennan-Wu/AISPFS

• Attentive Kernel and Robotic Information Gathering (Weizhe Chen) Code for robotic information gathering that maps spatial phenomena using point measurements, including our Attentive Kernel that facilitates calibrated uncertainty estimates and effective data collection is available through Weizhe Chen's github site. For context and description please see:
  AK: Attentive Kernel for Information Gathering , RSS 2022.

  Download code through:  https://github.com/weizhe-chen/attentive_kernels and https://github.com/weizhe-chen/pypolo

• DLM for sGP (Yadi Wei) Code for machine learning using sparse Gaussian Processes with our new method, direct loss minimization, is provided through Yadi Wei's github site. For context and description please see:
  Direct Loss Minimization for Sparse Gaussian Processes , AISTATS, 2021.

  Download code through: https://github.com/weiyadi/dlm_sgp

• SNAP (Hao Cui) Code for POMDP planning in large factored state, action and observation spaces is provided through Hao Cui's github site. For context and description please see:
  Sampling Networks and Aggregate Simulation for Online POMDP Planning, NeurIPS, 2019.

  Download code through: https://github.com/hcui01/SNAP

• SOGBOFA (Hao Cui) Code for stochastic planning in large factored state and action spaces is provided through Hao Cui's github site. For context and description please see:
  Stochastic Planning with Lifted Symbolic Trajectory Optimization , ICAPS 2019.
  Online Symbolic Gradient-Based Optimization for Factored Action MDPs , IJCAI, 2016.

  Download code through: https://github.com/hcui01/SOGBOFA

• AGS (Hao Cui) Code for solving Marginal MAP problems - max_M sum_S f(M,S,E) for max variables M, sum variables S, and evidence variables E, where f() is a Bayesian network with binary variables - is provided through Hao Cui's github site. The algorithm is the Algebraic Gradient based Solver. For context and description please see:
  From Stochastic Planning to Marginal MAP, NeurIPS 2018.
  

  Download code through: https://github.com/hcui01/AGS

• Variational inference in Bayesian latent Gaussian models (Rishit Sheth) Code for efficient supervised learning in the extended latent Gaussian model family via variational inference is provided through Rishit Sheth's Github site.
  For context and description please see: A Fixed-Point Operator for Inference in Variational Bayesian Latent Gaussian Models , AISTATS, 2016.

  Download code through:: https://github.com/rsheth80

• Sparse Gaussian Processes (Rishit Sheth) Code for efficient supervised learning with Gaussian Process models with arbitrary likelihood functions, through a sparse variational approximation is provided through Rishit Sheth's Github site.
  For context and description please see: Sparse variational inference for generalized Gaussian process models, ICML 2015.

  Download code through: https://github.com/rsheth80

• The OGLE II dataset (Yuyang Wang) The data in this archive includes the "OGLE II" dataset as used by the machine learning group at Tufts University. The dataset includes time series of light measurements from 3 type of periodic variable stars from the Optical Gravitational Lensing Experiment (OGLE) survey. The data includes a total of 14087 time series with (3425,3390,7272) in the categories (CEPH, RRL, EB).

  The data was generated and kindly provided by other researchers. Please see the official OGLE site for more information on the survey, data and discoveries. Various queries on this and other astronomy data can also be made there as well as at the visier site and Harvard time series site. If using this data please cite the original work (Szymanski, 2005, Acta Astron., 55, 43 and Udalski, Kubiak and Szymanski, 1997, Acta Astron., 47, 319.) as suggested on the OGLE site.

  Our group has made use of this data in machine learning research for classification, probabilistic modeling and period detection.

  We provide the OGLE-II dataset in order to make it more readily accessible to machine learning researchers. Toward that we packaged three versions of the data, as linked below. Please consult our papers for more information about the data, its processing and experiments.
  (1) The raw data: the original time series that are measured at irregular time points and are not folded. We also provide the folded versions, as well as the known period (and other properties) as found by the OGLE project: ogle2full.tar.gz
  (2) A processed form of the data: each time series is folded according to its known period, and then re-sampled via interpolation at 50 regular sampling time points. Two versions are provided, the time series "as is" and after "universal phasing". This form of the data can be simply treated as a point in 50-D Euclidean space and used directly by machine learning algorithms - providing an easy starting point to study the data. ogle50.libsvm and upogle50.libsvm

• FODD-Planner (by Saket Joshi) is a Prolog based software for stochastic planning problems based on the FODD representation developed by our group. The code implements symbolic dynamic programming using our representation of first order decision diagrams.
  Two systems are available under this heading: FODD and GFODD
  
  The most recent version of the FODD based system (includes code for JAIR2011, ICRA2012 and parallelization) is hosted as a google code project at: foddplanner google code site
  
  Older version of the code that can be used to reproduce the results from our JAIR 2011 paper: DISTRIB/foddplanner-jair11.tgz
  
  The most recent version of the GFODD based system (includes code for ECML2013 and more) is hosted as a google code project at: gfood planner google code site

• LogAn-H (by Marta Arias, Jerome Maloberti) is a system for learning function free Horn expressions. It is based on provably correct algorithms for learning with queries. More information can be found in:
  
  Theoretical background (correctness and complexity proofs)
  Paper describing the system and some experiments
  An Example Run
  
  Two variants of the system and algorithms have been implemented (in Prolog and C) as described in the papers.

• The L2Act system is a learning system that takes as input a description of an AI planning domain and a set of solved instances from that domain (the examples), and induces a set of rules ordered as a decision list that can be used as a reactive planner in this domain. More information can be found in:
  
  L2Act user manual
  Paper describing experiments with the system
  Paper describing relevant theoretical results