Microsoft Research has unveiled Icecaps, a new open source solution for neural conversational networks. The toolkit leverages multitask learning to improve conversation AI systems, such as giving them multiple personas.
Icecaps is an acronym that means Intelligent Conversation Engine: Code and Pre-trained Systems, uses both personality embeddings and word embeddings. By combining the two, the toolkit can personalize AI personas.
By using the toolkit, developers can give their AI assistants the ability to speak in different ways. For example, the assistant could tailor its speech and personality to individual users or to function in certain situations.
“Several of these tools were driven by recent work done here at Microsoft Research, including personalization embeddings, maximum mutual information-based decoding, knowledge grounding, and an approach for enforcing more structure on shared feature representations to encourage more diverse and relevant responses,” says Microsoft researcher Vighnesh Leonardo Shiv.
Microsoft says there are pre-trained models that developers can use or adapt for their assistants. These models will be released in the coming months.
“We had hoped to include these systems with Icecaps at launch. However, given that these systems may produce toxic responses in some contexts, we have decided to explore improved content-filtering techniques before releasing these models to the public,” the Icecaps GitHub page reads.
Microsoft used the TensorFlow machine learning framework to build the Icecaps library. Elsewhere, SpaceFusion is also integrated, which allows efficiency improvements in multi-task machine learning scenarios.
“Icecaps enables multi-task learning by representing most models as chains of components and allowing researchers and developers to build arbitrarily complex configurations of models with shared components. Flexible multi-task training schedules are also supported, allowing users to alter how tasks are weighted over the course of training.”