Author name cluster

Hung H. Bui

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

6 papers

2 author rows

ICML Conference 2020 Conference Paper

Collapsed Amortized Variational Inference for Switching Nonlinear Dynamical Systems

Zhe Dong
Bryan A. Seybold
Kevin Murphy 0002
Hung H. Bui

We propose an efficient inference method for switching nonlinear dynamical systems. The key idea is to learn an inference network which can be used as a proposal distribution for the continuous latent variables, while performing exact marginalization of the discrete latent variables. This allows us to use the reparameterization trick, and apply end-to-end training with stochastic gradient descent. We show that the proposed method can successfully segment time series data, including videos and 3D human pose, into meaningful “regimes” by using the piece-wise nonlinear dynamics.

Details

ICML Conference 2020 Conference Paper

Predictive Coding for Locally-Linear Control

Rui Shu
Tung Nguyen
Yinlam Chow
Tuan Pham
Khoat Than
Mohammad Ghavamzadeh
Stefano Ermon
Hung H. Bui

High-dimensional observations and unknown dynamics are major challenges when applying optimal control to many real-world decision making tasks. The Learning Controllable Embedding (LCE) framework addresses these challenges by embedding the observations into a lower dimensional latent space, estimating the latent dynamics, and then performing control directly in the latent space. To ensure the learned latent dynamics are predictive of next-observations, all existing LCE approaches decode back into the observation space and explicitly perform next-observation prediction—a challenging high-dimensional task that furthermore introduces a large number of nuisance parameters (i. e. , the decoder) which are discarded during control. In this paper, we propose a novel information-theoretic LCE approach and show theoretically that explicit next-observation prediction can be replaced with predictive coding. We then use predictive coding to develop a decoder-free LCE model whose latent dynamics are amenable to locally-linear control. Extensive experiments on benchmark tasks show that our model reliably learns a controllable latent space that leads to superior performance when compared with state-of-the-art LCE baselines.

Details

KER Journal 2010 Journal Article

Efficient variable elimination for semi-structured simple temporal networks with continuous domains

Neil Yorke-Smith
Hung H. Bui

Abstract The Simple Temporal Network (STN) is a widely used framework for reasoning about quantitative temporal constraints over variables with continuous or discrete domains. The inference tasks of determining consistency and deriving the minimal network are traditionally achieved by graph algorithms (e.g. Floyd-Warshall, Johnson) or by iteration of narrowing operators (e.g. ΔSTP). None of these methods exploits effectively the tree-decomposition structure of the constraint graph of an STN. Methods based on variable elimination (e.g. adaptive consistency) can exploit this structure, but have not been applied to STNs as far as they could, in part because it is unclear how to efficiently pass the ‘messages’ over continuous domains. We show that for an STN, these messages can be represented compactly as sub-STNs. We then present an efficient message-passing scheme for computing the minimal constraints of an STN. Analysis of this algorithm, Prop-STP, brings formal explanation of the performance of the existing STN solvers ΔSTP and SR-PC. Empirical results validate the efficiency of Prop-STP, demonstrating performance comparable to ΔSTP, in cases where the constraint graph is known to have small tree width, such as those that arise during Hierarchical Task Network planning.

Details DOI

AAAI Conference 2008 Conference Paper

The Hidden Permutation Model and Location-Based Activity Recognition

Hung H. Bui
Svetha Venkatesh

Permutation modeling is challenging because of the combinatorial nature of the problem. However, such modeling is often required in many real-world applications, including activity recognition where subactivities are often permuted and partially ordered. This paper introduces a novel Hidden Permutation Model (HPM) that can learn the partial ordering constraints in permuted state sequences. The HPM is parameterized as an exponential family distribution and is flexible so that it can encode constraints via different feature functions. A chain-flipping Metropolis-Hastings Markov chain Monte Carlo (MCMC) is employed for inference to overcome the O(n!) complexity. Gradient-based maximum likelihood parameter learning is presented for two cases when the permutation is known and when it is hidden. The HPM is evaluated using both simulated and real data from a location-based activity recognition domain. Experimental results indicate that the HPM performs far better than other baseline models, including the naive Bayes classifier, the HMM classifier, and Kirshner’s multinomial permutation model. Our presented HPM is generic and can potentially be utilized in any problem where the modeling of permuted states from noisy data is needed.

PDF Details

AAAI Conference 2004 Conference Paper

Hierarchical Hidden Markov Models with General State Hierarchy

Hung H. Bui

The hierarchical hidden Markov model (HHMM) is an extension of the hidden Markov model to include a hierarchy of the hidden states. This form of hierarchical modeling has been found useful in applications such as handwritten character recognition, behavior recognition, video indexing, and text retrieval. Nevertheless, the state hierarchy in the original HHMM is restricted to a tree structure. This prohibits two different states from having the same child, and thus does not allow for sharing of common substructures in the model. In this paper, we present a general HHMM in which the state hierarchy can be a lattice allowing arbitrary sharing of substructures. Furthermore, we provide a method for numerical scaling to avoid underflow, an important issue in dealing with long observation sequences. We demonstrate the working of our method in a simulated environment where a hierarchical behavioral model is automatically learned and later used for recognition.

PDF Details

AAAI Conference 2000 Conference Paper

On the Recognition of Abstract Markov Policies

Hung H. Bui
and Geoff West

Abstraction plays an essential role in the way the agents plan their behaviours, especially to reduce the computational complexity of planning in large domains. However, the effects of abstraction in the inverse process – plan recognition – are unclear. In this paper, we present a method for recognising the agent’s behaviour in noisy and uncertain domains, and across multiple levels of abstraction. We use the concept of abstract Markov policies in abstract probabilistic planning as the model of the agent’s behaviours and employ probabilistic inference in Dynamic Bayesian Networks (DBN) to infer the correct policy from a sequence of observations. When the states are fully observable, we show that for a broad and often-used class of abstract policies, the complexity of policy recognition scales well with the number of abstraction levels in the policy hierarchy. For the partially observable case, we derive an efficient hybrid inference scheme on the corresponding DBN to overcome the exponential complexity.

PDF Details