Microsoft Open-Sources Its Intelligent Conversation Engine - Icecaps

Icecaps tools are driven by up-to-date work at Microsoft Research.

Recently, Microsoft announced the open-sourcing of its Intelligent Conversation Engine - Microsoft Icecaps. Icecaps toolkit allows developers to imbue their chatbots with different personas as well as but to consolidate other natural language processing features that emphasize conversation modeling.
 
The solution offers an array of capabilities from recent conversation modeling literature. Many of these tools are driven by up-to-date work at Microsoft Research, such as personalization embeddings, maximum mutual information–based decoding and knowledge grounding. These tools, according to the company, come with an approach for enforcing more structure on shared feature representations to encourage more diverse and relevant responses.
 
Microsoft said that its library leverages TensorFlow in a modular framework designed to make it easy for users to construct sophisticated training configurations using multi-task learning. And soon the team will equip Icecaps with such pre-trained conversational models that you can either use directly out of the box or quickly adapt to new scenarios by bootstrapping your own systems.
 
According to the company, Icecaps, in order to enable multi-task learning, represents most models as chains of components and allows you to build arbitrarily complex configurations of models with shared components. And, it also supports flexible multi-task training schedules which allows users to alter how tasks are weighted over the course of training.
 
ICECAPS
Source: Microsoft 
 
Icecaps also implements SpaceFusion, which is a specific multi-task learning paradigm originally created to jointly optimize for diversity and relevance of generated responses.
 
 
 
Source: Microsoft