AI Planning

Overview

We create new planning algorithms and develop applications of planning to business areas such as risk management, defense, dialogue management, healthcare, cybersecurity, analytics and public transportation.

What is AI Planning

Planning is a long-standing sub-area of Artificial Intelligence (AI). Planning is the task of finding a procedural course of action for a declaratively described system to reach its goals while optimizing overall performance measures. Automated planners find the transformations to apply in each given state out of the possible transformations for that state. In contrast to the classification problem, planners provide guarantees on the solution quality.

Why is it Important: Planning Applications in Industry

Automation is an emerging trend that requires efficient automated planning
Many applications of planning in industry (e.g. robots and autonomous systems, cognitive assistants, cyber security, service composition)

How to Spot a Planning Problem

Declarative

You want to find a procedural course of action for a declaratively described system to reach its goals while optimizing overall performance measures.

Domain Knowledge can be elicited or learned over time

Existing domain knowledge can/should be exploited for building the model
Human involvement controllable. Humans build the model and can contribute to the solution by introducing knowledge.

Favor consistency over learning transient behaviors

There is a structure of the problem that cannot be learned just training
When no large training data is available
Changes in the problem can make previous data irrelevant

Advantages of AI Planning Techniques

When explainability is desired

When you want to be able to explain why a particular course of action was chosen
Assignment of responsibility/blame is essential for automation of processes (e.g., autonomous driving, medical expert systems)

Rapid prototyping: short time to solution

Variety of of-the-shelf planners available both IBM proprietary and open-source

Your problem is frequently changing, even small changes.

No need to change the solution, only tweak the model

Success Stories: When Planning Meets DL

In many real life applications, there is a structure of the problem that cannot be learned with DL (there are just not enough examples). Solving optimization problems with learning is hard, but integrating planning techniques with heuristic guidance learned by DL will result in the most famous success stories of AI to day.

GO player AlphaGO uses planning (monte-carlo tree search) with deep learning (heuristic guidance) to select the next move
Cognitive assistant Viv (Samsung) uses knowledge graph, planning, and deep learning to answer complicated queries

Example AI Planning Projects in IBM

Current Projects

IBM Research Scenario Planning Advisor
Mercury Planner: Award-winning open-source
Top-k Planner: State of the art Top-k planner integrating the K* algorithm into Fast Downward
DRL-CPLAN: Optimal non-deterministic planning based on state-of-the-art efficient memory-limited AND/OR graph search.

Past Projects

DOCIT: Multi-modal journey planning with contingencies for missed connections
Planner4J: A framework of Java-based planners for enterprise applications
Synthy: An end to end approach for composition and adaptation of web services, and relevant to other component technologies
MARIO: An automated semantics-based flow composer with faceted goal navigation
LTS++: Planning-based Hypothesis Generation with LTS++