Modes of Machine Learning Sample Clauses

Modes of Machine Learning. In addition, algorithms can be trained ‘online’ or ‘offline’, using sequential or simultaneous (batch) data inputs, respectively.192 Online learning sees the program learning incrementally from new external data, as and when these are collected by its sensors.193 This enables the system to exploit growing volumes of data, where these are unknown, or represent changes to existing datasets.194 In a LAWS context, this would mean learning from interactions in a live battlefield, which would enable the system to: • adapt to new backgrounds and climatic conditions; • refine its warfighting tactics, for greater targeting accuracy and precision; • keep pace with changing enemy tactics in real-time, for example, by developing more effective evasive/defensive manoeuvres or new methods of attack; and • generalise new target sets that may have been missed during the formal targeting process. Offline learning is where the program is restricted to its original training data.195 For a LAWS, this largely refers to laboratory training, which takes place during product development, and will be subject to human scrutiny via verification and validation procedures before the algorithmic changes are finally integrated into the system. It can be expected that in a warfighting context, offline learning will be preferred for the greater predictability and human control that it affords. However, as online data enable algorithms to remain up-to-date and combat-effective, they are still likely to be recorded by a LAWS on deployment, before being reviewed by human programmers and selectively uploaded during system updates. It is also possible that limited forms of online learning will be allowed, for example, to refine the machine’s targeting functions but without undertaking whole new tasks or methods of warfare that were not anticipated by deploying commanders. 192 Xxxxxx Xxxxxx and Xxxxxxxx X. Xxxx (eds.), Encyclopedia of Machine Learning (Springer, 2010), 74, 736-43; Xxx Xxxxxxx, Elements of Machine Learning (Xxxxxx Xxxxxxxx, 1996), 9. 193 Xxxxxxx and Xxxxxx (n 160), 752-53. 194 Ibid. 195 Xxxxxx and Xxxx (n 192); Xxxxxxx (n 192).