Evaluating and improving a probabilistic threat assessment algorithm

在军事、安全和监视行动中，威胁评估包括分析当前情况并预测当前情况下哪些特工有意图和能力危及我们的任务目标。潜在的认知复杂性需要自动化算法来支持人类操作员分析威胁。设计这类算法的关键挑战之一是能够基于对其行为影响的观察来自动推断一个或多个观察者的意图、目标或计划。当今大多数有前途的技术都依赖于概率推理。NSERC Engage项目的目标是首先测试和评估Menya Solutions Inc.开发的威胁评估算法的性能，该算法使用的场景比Menya到目前为止考虑的场景更复杂。其次，我们打算减少开发该算法所需的威胁行为模型作为输入所需的工作量，首先通过探索使用深度学习来从模拟中学习行为模型，然后通过研究基于一种称为反向规划的方法的技术。

In military, security and surveillance operations, threat assessment consists of analyzing the current situations and predicting which agents in the current situation have the intent and capability to endanger our mission goals. The underlying cognitive complexity requires automated algorithms to support human operators analyzing the threats. One of the key challenges in designing such algorithms is being able to automatically infer the intent, goal or plan of an observant agent or groups of agents based on observations of the effects of their actions. Most of today's promising techniques rely on probabilistic reasoning. The objective of this NSERC Engage project is first to experiment and evaluate the performance of an algorithm for threat assessment developed by Menya Solutions Inc., using scenarios that are more complex than those considered by Menya so far. Secondly, we intend to reduce the effort required in developing the threat behaviour models required by this algorithm as input, first by exploring the use of deep learning to learn behaviour models from simulations and then by studying a technique based on a method called inverse-planning.