Building the next-generation Conversational AI with Python and Deep Learning
summary
Shows building QA and dialog systems in Python using deep learning with ParlAI (20+ tasks, Mechanical Turk integration) and SQuAD-style training; claims an end-to-end trainable model with character+word embeddings and context layers that extracts short and long answers from passages (demo on Melbourne wiki). Runs multitask training to reduce overfitting, uses a 90% confidence threshold to avoid bad replies, and requires retraining on policy changes. Two enterprise chatbots launch in production; batch retrains take about a week for typical datasets.
AI-written summary · about
chapters (7)
- 0:00Start
- 5:09Glorified IVR System
- 6:33Dialog Research Software Platform
- 10:15Multi-Task Learning (MTL)
- 12:21Datasets
- 13:285 Categories
- 17:42Deep Reading Comprehension Demo
related talks
30mConversational AI: Building clever chatbots
38mBuilding your own Multiskill AI Assistant with DeepPavlov
31mMachine Learning-Enabled Chatbots
41mConversational Agents at Scale: Retrieval and Generative approaches
01h 31mBuilding a ChatBot with Python, NLTK and scikit
47mTesting Conversational AI
01h 38mConversational AI with Rasa Core & NLU
01h 29mGoing Beyond “Sorry, I Didn’t Get That”: Building AI Assistants that Scale...