AI Neural Network and NLP In Chatbots

After confirming the topics with available speakers we now have more details of this meetup.

We are delighted to have three females in this event, and again with Accenture, to give the speech associated their study, project and ideas in deep learning. Join us from 18:30 on 25th of June. This time we have the speakers from Insight and Adapt Labs, also from Accenture R & D team, please see the below with the brief of the talks.

The same information will also be published on our web pages
http://www.deeplearning.ie

[IMPORTANT]

Bring your ID when you come to event. We will need to see it before granting you entrance.

Leave us comment if you want secure your attendance.

*** AGENDA ***

[18:15] Registration & Social Networking, enjoy the Pizza and drinks

[19:00] Welcome and opening by Accenture the Dock Lead for Analytics Rory Timlin & EI Studio Lead Stephen Redmond

[19:15 - 19:40] Dr. Conor O'Shea, Developer and Co-funder of Darwin & Goliath

Topic
Using Machine Learning to predict the best recommendation approach for a given user/context.

Brief
We have found that recommendations algorithms vary a lot in their effectiveness in different contexts such as domiain, user demographics, and even time of day. Darwin & Goliath uses a unique approach to solve this issue. We use machine learning to predict which recommendation algorithm will perform best for the current context, and use that algorithm to generate the recommendations. Using this adaptable approach we hope to be able to provide Recommendations-as-a-Service to a very broad range of domains including fintech, legal, and health as well as the more common e-commerce and online content use cases. Currently we use a Random Forest algorithm to select a recommendation algorithm on a per-request basis. We would like to try a neural net in its place in future to see if it can produce better results.

[19:40-20.05] PHD Miss Zoey An from UCD Insight

Topic
Automatic Geological Fault Interpretation And Modelling

Brief
Fault interpretation is critical in finding the presence of natural resources, like oil, gas, however, the manual label process is time-consuming. In the last decade, deep convolutional neural networks have successfully applied in multi-discipline and shown very high performance on various image processing tasks. Therefore, we purpose using deep convolution neural networks (e.g. Unet) to interpret faults from the perspective of semantic image segmentation.

In this talk, I will first introduce some background of geology faults. Then, identify current challenges. After that, I will explain our baseline solution followed by some possible future works.

[20:05 - 20:30] Dr. Marissa Zhou Data Science Manager and Dr. Alana Rivera Ingraham, both from Accenture

Topic
Deep Learning in Natural Language Understanding used in Chatbots

Brief
An ongoing successful also interested project from R & D team. This talk will cover how Deep Learning is applied in Natural Language Understanding when building Chatbots

[20:30 - 20:55] Patrick Moreau from Accenture.

Topics
Latent Space Exploration for New Molecule Generation

Brief
Introduction to autoencoders.
From vanilla to variational autoencoders.
Methods to explore the latent space.
An example use-case from an internal project (drug discovery).