Scribd
Upload
73 views

Learning AI Development With UX

This document summarizes a meetup on AI development with a UX focus. The agenda includes discussing the machine learning mindset of identifying good problems, hands-on exercises for building datasets and text classification models, and challenges of real-world AI development like quantifying goals, obtaining labeled data, model serving, and developer productivity. The presenters are an NLP engineer and software engineer who provide an overview of machine learning concepts and lead activities on preparing data and basic neural networks for text classification.

Uploaded by

A Rao
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
73 views

Learning AI Development With UX

This document summarizes a meetup on AI development with a UX focus. The agenda includes discussing the machine learning mindset of identifying good problems, hands-on exercises for building datasets and text classification models, and challenges of real-world AI development like quantifying goals, obtaining labeled data, model serving, and developer productivity. The presenters are an NLP engineer and software engineer who provide an overview of machine learning concepts and lead activities on preparing data and basic neural networks for text classification.

Uploaded by

A Rao
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 41

December Meetup

Learning AI Development
with UX
Who are we?

Arjun Rao Anisha Mascarenhas


NLP Engineer at Stride.ai Inc Software Engineer at LinkedIn
Agenda
1 The ML Mindset: Identifying good problems for AI

2 Hands on: Gather your ingredients for ML

3 A Text Classification Cookbook

4 Wrapping Up: Quirks of Real World AI


Development
The ML Mindset
What is AI?
At the core of every computer program there is a mathematical function at work. It could be

as simple as computing the interest on an outstanding loan or as complex as flying an aircraft

on autopilot. Artificial Intelligence, or AI, is a generic name for a computer program whose

core mathematical function has been created (almost) automatically; and Machine Learning,

or ML, refers to a collection of techniques which offer ways of creating AI.

Namit Chaturvedi
(PhD in theoretical computer science,
Applied Research Engineer at LinkedIn)
AI can only be as good as the examples and techniques used to train it
Thinking about a problem from a ML Perspective: From programs to experiments

Step Example

1. Set the research goal. I want to predict how heavy traffic will be on a given day.

2. Make a hypothesis. I think the weather forecast is an informative signal.

3. Collect the data. Collect historical traffic data and weather on each day.

4. Test your hypothesis. Train a model using this data.

5. Analyze your results. Is this model better than existing systems?

6. Reach a conclusion. I should (not) use this model to make predictions, because of X, Y, and Z.

7. Refine hypothesis and repeat. Time of year could be a helpful signal.


Identifying good problems for ML

Start with the problem, and not the solution


Identifying good problems for ML: Aim to make decisions, not just predictions.

"I trained a model that predicts the

probability that someone will want to

watch a video and still click "thumbs down"

on youtube!"
Hands On
Building a good dataset
Link to Dcult -
https://tinyurl.com/altimetrik-dcult

https://dcultaltimetrik.com
d
Link to Colab - https://goo.gl/tVgzsQ
Preparing your dataset
Preparing your dataset
Splitting your dataset:
Transforming data to features:
Example for dealing with Categorical Data
Transforming data to features:
Building a vocabulary
Transforming data to features:
Mapping categories to feature vectors
Transforming data to features:
Mapping categories to continuous valued feature vectors
Transforming data to features:
Mapping categories to continuous valued feature vectors
Visualizing Embeddings
Questions?
Hands On
A Text Classification
Cookbook
A note on neural networks
A note on neural networks: A single neuron
A note on neural networks: The activation function f
A note on neural networks: Activation functions
Term Frequency over Inverse Document Frequency: TF-IDF
Wrapping Up
Quirks of Real World AI Development
Quirks of Real World AI Development

● From unquantifiable goals to measurable metrics


● Getting a good labelled dataset
● Serving an ML Model
● Developer productivity
From unquantifiable goals to measurable metrics

● Finding the right metric


● Measuring success on real data
● Metrics may be influenced by other changes
Getting a good labelled dataset

● Direct or Derived labels


● Finding a good sample to label
● Wrong labels in large datasets
● Bias in data
Serving an ML Model

● Kind of model and features to use


- Offline
- Online
- Nearline
- On device
Serving an ML Model: A simple scenario
Developer Productivity

● Versioning of models and datasets


● Searchable and reproducible experiments
● Monitoring performance, A/B testing,
Debugging
Any Questions?
Slides, Code and Links
can be found at github.com/arjun-rao/talks/

Contact Us:
LinkedIn/Twitter: arjunra0
Linkedin:anishamascarenhas
Twitter: anisham197
References & Useful links
● https://ai.google/education
● https://developers.google.com/machine-learning/
● https://research.fb.com/the-facebook-field-guide-to-machine-le
arning-video-series/
● https://hackernoon.com/a-guide-to-scaling-machine-learning-m
odels-in-production-aa8831163846

You might also like