Media Summary: When you're uncertain about a prediction it's probably best not to immediately automate it. In this video, we'll discuss a What might an assistant do if it sees a text from a language that it isn't trained on? It might make assumptions because it's unlike ... In the previous video, we've started measuring bias in word embeddings. In this video, we will attempt to remove some of this bias ...
Rasa Algorithm Whiteboard Fallback Detection - Detailed Analysis & Overview
When you're uncertain about a prediction it's probably best not to immediately automate it. In this video, we'll discuss a What might an assistant do if it sees a text from a language that it isn't trained on? It might make assumptions because it's unlike ... In the previous video, we've started measuring bias in word embeddings. In this video, we will attempt to remove some of this bias ... In the previous video on bias, we showed that you should be careful with debiasing techniques. In this video, we zoom in on why ... In the age of deep learning and transformers, rule-based systems can still be a great idea. In this video, we hope to demonstrate ... The way the response selector is implemented in
What if we design a policy mechanism that doesn't predict the next action to take, but instead predicts when something ... In this video, we will discuss some technical details that are in the UnexpecTEDIntentPolicy. The UnexpecTEDIntentPolicy really is ... This is the first video on attention mechanisms. We'll start with self attention and end with transformers. We're going at it step by ... In this video we'd like to demonstrate an new tool that we've open-sourced. It's called "whatlies" and the goal of the package is to ... Bad labels make your benchmarks unreliable, so how might we find some? In this video, we'll try out one trick and we'll see if it ... In this video, we'll try to find unexpected intents using the brand new UnexpectedIntentPolicy. It helps to have seen the previous ...
In the previous videos, we've started measuring bias in word embedding and tried to counteract the bias with a linear algebra ... Typos are a common issue for NLU models. There are some tweaks you might do to your pipeline to combat them, but maybe it ... Text translation is hard, especially when you've only gotten a small amount of data to work with. But still ... we may be interested in ... In this video, we'll highlight a qualitative argument of why you may not need to worry about pre-trained embeddings too much. When people refer to a Machine Learning backend, they typically call it a "pipeline". A pipeline is a nice mental concept, and ... This is the second video on attention mechanisms. In the previous video we introduced self attention and in this video we're going ...
