Skip to main content

3 posts tagged with "validation"

View All Tags

Fixing Labelling Bias With Re:infer

· 13 min read
Rob Sumner
Machine Learning Research Engineer

Re:infer uses machine learning models to identify patterns in communications data like emails, chats and calls. Models extrapolate these patterns to make predictions for similar data in the future, driving downstream processes like automations and analytics.

For this approach to work, the data used to train a model needs to be representative of the communications it will make predictions on. When this is not the case, models will make mistakes that can seriously impact the performance of systems which rely on accurate predictions.

To help users build robust, well-performing models, we built a tool to ensure data used for training always matches the user’s target task. In this blog post we discuss how this tool works, and some of the problems we tackled during its development.

It's All in the Numbers - Assessing Model Performance With Metrics

· 10 min read
Rob Sumner
Machine Learning Research Engineer

When building and training a machine learning model, understanding performance is essential. Depending on the training data and task, even the most advanced model can produce incorrect predictions, resulting in misleading analytics or faulty automation flows.

Manually wading through examples to check model predictions is impractical, especially for datasets with millions of data points. Instead, Re:infer continuously computes and displays multiple metrics to help to analyse models and spot failures.

However, under some conditions metrics can behave unexpectedly. In this blog post we discuss some of the problems that arise when using metrics, and some of the solutions Re:infer uses to simplify the process.

Why Model Validation Is Important

· 7 min read
Rob Sumner
Machine Learning Research Engineer

Re:infer is a conversational data intelligence platform that enables users to discover, measure, and automate processes hidden in their communication channels.

Typical channels include emails, tickets, chats, and calls. Conversations in these domains are complex and nuanced. As a result, generic machine learning models perform poorly. Instead, Re:infer allows anyone to create custom models with little effort. No technical knowledge required.

This is an incredibly powerful approach. Models can learn complex patterns and make predictions on unseen data, just like humans. Machine learning models have even out-performed humans on some natural language tasks.

But like humans, machine learning models can also make mistakes. Estimating how often a model will be wrong is crucial for any real-world application of machine learning. Of equal importance is presenting this intuitively and highlighting the best actions to improve a model. Re:infer uses model validation to accomplish all these tasks.