Keiji Kanazawa

IJCAI Conference 1995 Conference Paper

Local learning in probabilistic networks with

• hidden variables Stuart Russell John Binder Daphne Roller
• Keiji Kanazawa

UAI Conference 1995 Conference Paper

Stochastic simulation algorithms for dynamic probabilistic networks

• Keiji Kanazawa
• Daphne Koller
• Stuart Russell 0001

Stochastic simulation algorithms such as likelihood weighting often give fast, accurate approximations to posterior probabilities in probabilistic networks, and are the methods of choice for very large networks. Unfortunately, the special characteristics of dynamic probabilistic networks (DPNs), which are used to represent stochastic temporal processes, mean that standard simulation algorithms perform very poorly. In essence, the simulation trials diverge further and further from reality as the process is observed over time. In this paper, we present simulation algorithms that use the evidence observed at each time step to push the set of trials back towards reality. The first algorithm, �evidence reversal� (ER) restructures each time slice of the DPN so that the evidence nodes for the slice become ancestors of the state variables. The second algorithm, called �survival of the fittestz� sampling (SOF), �repopulates� the set of trials at each time step using a stochastic reproduction rate weighted by the likelihood of the evidence according to each trial. We compare the performance of each algorithm with likelihood weighting on the original network, and also investigate the benefits of combining the ER and SOF methods. The ER/SOF combination appears to maintain bounded error independent of the number of time steps in the simulation.

AAAI Conference 1994 Conference Paper

Sensible Decisions: Toward a Theory of Decision-Theoretic Information Invariants

• Keiji Kanazawa

We propose a decision-theoretic notion of invariance in bounded rational decision making. We show how optimal decision making in sensory robotics can be approximately preserved under transformations of the decision rule. In particular, we present a decision theoretic analysis of the use of visual routines in action arbitration in real-time robot soccer. In this domain, stochastic dominance, and therefore decisions, can be sensed approximately from the environment, and we exploit this in our decision making.

AAAI Conference 1991 Conference Paper

A Logic and Time Nets for Probabilistic Inference

• Keiji Kanazawa

In this paper, we show a new approach for reasoning about time and probability that combines a formal declarative language with a graph representation of systems of random variables for making inferences. First, we provide a continuous-time logic for expressing knowledge about time and probability. Then, we introduce the time net, a kind of Bayesian network for supporting inference with statements in the logic. Time nets encode the probability of facts and events over time. We provide a simulation algorithm to compute probabilities for answering queries about a time net. Finally, we consider an incremental probabilistic temporal database based on the logic and time nets to support temporal reasoning and planning applications. The result is an approach that is semantically well-founded, expressive, and practical.

IJCAI Conference 1989 Conference Paper

A Model for Projection and Action

• Keiji Kanazawa
• Thomas Dean

In designing autonomous agents that deal competently with issues involving time and space, there is a tradeoff to be made between guaranteed response-time reactions on the one hand, and flexibility and expressiveness on the other. We propose a model of action with probabilistic reasoning and decision analytic evaluation for use in a layered control architecture. Our model is well suited to tasks that require reasoning about the interaction of behaviors and events in a fixed temporal horizon. Decisions are continuously reevaluated, so that there is no problem with plans becoming obsolete as new information becomes available. In this paper, we are particularly interested in the tradeoffs required to guarantee a fixed reponse time in reasoning about nondeterministic cause-andeflect relationships. By exploiting approximate decision making processes, we are able to trade accuracy in our predictions for speed in decision making in order to improve expected performance in dynamic situations.

UAI Conference 1988 Conference Paper

Probabilistic causal reasoning

• Thomas L. Dean
• Keiji Kanazawa