DEV Community: Michael Lyam

How does Google AutoML Works?

Michael Lyam — Mon, 23 Mar 2020 10:39:38 +0000

AutoML is becoming an important part of machine learning for the futuristic ML/AL engineers.

Let us assume that business ‘A’ identified a problem and called their Artificial Intelligence or Machine Learning engineers’ team. The team then gets briefed about the situation and were tasked to arrive at a predictive analytics solution.

The traditional workflow involves different team collaborating together to arrive at a solution.

The workflow is complex involving –

This is what traditionally the team can do –

It is a long procedure and takes time.This face of ML-based solutions is changed by Google’s AutoML.

Where, the workflow is minimized as –

This is the basics of AutoML. The steps that take place between data acquisition and prediction is abstracted by the AutoML platform. The businesses can bring their dataset, identify labels, and set the button. The trained and optimized machine learning model predicts. In brief, most of the steps are handled behind the scene providing time and opportunity for businesses to stay focused on problem and solutions rather than process and workflow.

Many of the AutoML platforms support to export the trained model into mobile devices running in iOS and Android, enabling developers to integrate easily within their mobile devices. As they get exported into the Docker container, DevOps team can deploy them at scale and infer in production environments.

So, AutoML is promising for the non-tech companies to build ML applications and access the capabilities at lower costs. Isn’t it?

Google launched its Cloud AutoML in 2018. It has its proprietary algorithm. AutoML enables businesses to benefit from data-driven applications powered by statistical models. It can automate many of the tasks performed by data scientists.

Google uses premium Transfer Learning and Neural Architecture Search technology to conduct automation in machine learning through cloud.

A few of the libraries used for automating machine learning are as listed below.

• Eclipse Arbiter is a hyperparameter optimization library that automates hyperparameter tuning for deep neural net training.
• Featuretools automates engineering features from relational and transactional data
• Auto-sklearn is a replacement for scikit-learn estimators
• MLBox to support model stacking
• TPOT to find the ML pipelines that are best performing

Other libraries include Xcessive, Advisor, Hyperpot, Spearmint, RoBo, BayesianOptimization, Optunity, ATM, and HyberBand.

Let us understand, how Google’s AutoML is beneficial for the businesses while predicting solutions.

Benefits of using Google’s AutoML

A few of them are briefed here. They include –
• Runs repetitive tasks automatically with improved efficiency.
• Facilitates data scientists to focus on problems rather than models.
• Avoids potential manual errors.
• Allows everyone to use machine learning features.
• Reduces time to implement machine learning process.
• Organizations can build production-ready models quickly.
• Improved scalability as the model can get deployed for different use cases.

Moving forward, let us see a couple of real-world instances for AutoML.

1) Google AutoML to detect Pneumonia with chest X-ray images

Google Cloud AutoML Vision simplifies the creation of custom vision models to recognize images. It is used to develop medical image classification model that can detect pneumonia using chest X-ray images.

2) Restaurant location recognizing model

AutoML is used to identify the restaurant by looking at the image of the noodle bowl. The model is able to analyze minute details of the image and predict which restaurant it was made in.

There are several real-world examples that demonstrates the capabilities of AutoML. So, AutoML seems to be a promising solution for companies to bridge the talent gap in the data science industry.

It is an important part of machine learning for AL/ML engineers as AutoML happens to be the future of Artificial Intelligence.

What Are The Steps in AI Chatbot Development

Michael Lyam — Wed, 05 Feb 2020 06:18:09 +0000

Being a simple text-based solution to AI-powered conversational bots, businesses have come a long way to support and engage customers 24*7. Human-like chatbots simulate real-time conversations and streamline interactions to improve customer experience and increase online sales. On the other hand, it reduces operational costs too.

Well, Gartner recently reported that 85% of customer interactions will be managed without human in 2020.

Businesses use bots because they perform simple and structurally repetitive tasks at a much higher rate. For instance, a single bot can handle queries of hundreds of customers at the same time. Chatbots can be integrated to leading chat platforms like WhatsApp, Slack, Telegram, Facebook Messenger, Twilio, etc.

Chatbots are used in businesses to:

In brief, these chatbots manage the communication gateways of several business and companies like banks, telecom service, travel companies, e-commerce portals, drug manufacturer, insurers, etc.

There are two categories of chatbot –

(a) that runs on rules
The bots work on a specific set of rules. It fails to act when anything comes beyond its purview.

(b) that runs on machine learning
The following picture shows the working principle of chatbot using Artificial Intelligence.

As a first step, AI engineer has to understand the opportunity to build an AI-based chatbot. A few types of work can be automated while others need augmentation with Artificial Intelligence solutions. Further, AI engineer has to

• Identify the problem that bot would solve
• Design a conversational user flow
• Choose the platform where the bot will reside (messenger, Facebook, etc.)
• Set up server to run the bot

There are two main phases involved to build a chatbot, namely, conversation design and the construction of the bot.

1) Conversation design:

This requires the human element and AI engineer’s thinking and decision capability plays a major role here. This encompasses flow and scripting.

The AI engineer develops the chatbot using NLP and Machine Learning. He prepares the flow like

• What content the bot may provide
• What questions it may have to answer
• What actions it should take
• What the end user might ask
• When to redirect to a live agent

A successful AI chatbot breaks the user statements into context (user, time, profile), entities (objects of conversation), and intent (user wants). NLP systems use the variables and plan responses. So, an AI engineer considers all the entities and intents to come up with possible responses.

Further, the AI engineer continues with scripting, i.e., gives the bot a persona like personality, voice, and tone depending on market trend.

Let’s move on to the next phase.

2) Development of chatbot:

Some of the popularly used platforms to build a bot include Chatfuel, Botsify, Pandorabots. However, the best recommended option is to use frameworks. The popular frameworks include Dialog Flow, Microsoft Bot Framework, Facebook Bot Engine, Amazon Lex, IBM Watson Assistant, Aspect, etc.

Source: Chatbot Magazine

The third and last step involves testing of the chatbot.
Generally, AI engineers use testing tools and ready-made solutions like Selenium, Zypnos, or TestMyBot etc., by determining KPIs.

The secret to create a best chatbot is to put thought process and effort to construct the flow by considering the business goals and make it work using technology. So, it is essential to learn business skills and earn AI certification to gain competitive advantage in the AI field.

Golang Vs Python: What’s Your Best Pick For AI

Michael Lyam — Wed, 29 Jan 2020 05:58:19 +0000

Python programming language has been gaining popularity over the past few years.

Total programming languages worldwide is estimated to be between the range of 256 and above 8,000. However, every developer has their favorite language. But there are only a few numbers of programming languages that top the list when it comes to broad use across various domains and industries.

AI programming has gone far beyond one could speculate. Companies dealing with healthcare, real estate, or verticals like e-commerce are now adopting AI in their product development. Before getting into a conclusion, most development team decides which language would be the right pick for the project. Thus, many times choosing the language comes down to Golang and Python.

However, there are certain parameters to take heed when comparing which programming languages to choose from. Some of the parameters are:

1.Scalability

Golang

The very first fact that should click out mind should be, Google developed Golang in an attempt to help developers working in Google solve critical problems. This essentially means the tech giant has hundreds and thousands of developers working on the same software that is hosted on many clusters. Thus, solving problems at a larger scale.

Python

On the other hand, Python does face issues with concurrency but it can still implement parallelism via threads. This simply means that the application will split their task up into smaller subtasks that will be processed in a parallel, but in multiple CPUs and at the same time.

2.Performance

In terms of performance, both Python and Golang are excellent programming languages. But it is said that Google’s compiled language is much faster than the interpreted language such as Python. However, speed is just a primary concern for software development since both these languages are not in much demand on memory and CPU. Also, both Golang and Python follow different kinds of approaches while handling concurrency.

3.Applications

Golang

Golang has been used majorly for systems programming. And since the language supports concurrency it is ideal for cloud computing and cluster computing.

Python

For every specific programming language, there’s a purpose. In the same manner, Python has been widely used in the field of AI, deep learning, web development, and data analytics.

4.Code readability

Golang

Golang comes with strict rules when there’s the involvement of programming. It prevents from developing unnecessary libraries and frameworks. Thus, this gives better code readability by eliminating unnecessary creation of code. Although Python wins most of the developers’ choice of programming languages, Golang will surely triumph and take a good hit in AI.

Python

For instance, if you’re developing software for your organization or a client, it is evident you’ll be working with a team or perhaps several AI engineers. And at such instances, code readability becomes a huge factor that needs to be taken into consideration.

Although Python offers fabulous readability. However, at times doing things in ten different ways may lead to a lot of confusion.

5.Libraries

Golang

When Go was developed, Google ensures that the libraries should be a part of their inbuilt Go libraries. Although these libraries may not be as robust as Python, the usage of these libraries covers almost every aspect used for AI programming.

Python

Python has multiple libraries and frameworks and this is one of the biggest advantages. Flask and Django are the two most popular frameworks of Python. However, there are newer libraries that are specifically used in AI programming and machine learning.

• Keras and Tensorflow – building deep learning models
• NumPy – for data cleaning and data manipulation
• Scikit-learn – modeling
• Caffe – for image processing and classification
• PyTorch – training deep learning models
• Matplotlib – for data visualization
• Pandas – data manipulation and data analysis
• OpenCV- image processing

Summing up

Both Go and Python have their specific unique features, but it is for you to take the call and choose the language that best fits your preference. However, this also depends on the kind of project you’re working with depending on the end-goal.

Originally Published: Golang Vs Python: What’s Your Best Pick For AI