AI-based solutions have seen a huge rise in the last decade and are being deployed in more areas. Any solution design is built with the scope for AI. AI helps in improving the experience of the user and also acts on behalf of the user by providing valuable suggestions. We are at the initial stages of AI, and Machine Learning(ML) is the basic building block for AI.
Predominantly, big companies go for AI-First approach and it is the most widely adopted solution design. Some of the recent trends show the adoption of ‘AI-Mobile first approach’.
Answering the below question would help you understand the importance of ‘AI-Mobile first approach’-
1) Who is the end user of the solution?
2) What is the predominant form factor that the solution is going to run on?
3) Does the solution violate any privacy laws?
4) Do we have enough and right data for ML? and so on
What is machine learning? How is it different from automation?
There are many standard book definitions for ML. To me, machine learning is something that can be taught or trained to machines by utilizing the old data or experience using algorithms. As the size of the sample data grows, the more the machines can be trained with more accuracy. So how does this training happen? I would say it all started with telling machines with simple pre-written rules.
Let’s see with simple examples. If we want to understand shapes then initially our rules might have said –
-> If an object contains 3 sides then its triangle
-> If an object contains 4 sides then its rectangle
Now our algorithm(aka model) would be able to understand triangle and rectangle.
What if you want your model to understand pentagon, hexagon and more? Then you re-write your rules to include –
-> If an object contains 5 sides then its pentagon
-> If an object contains 6 sides then its hexagon
In this case, we keep on adding our rules. If you see the way we are adding rules we just automate the way we identify the shapes by manually adding rules. But the goal of machine learning is to automatically incorporate these rules in the form of models by classifying based on the training data.
What is the scope of ML in mobility?
Mobile solutions are not those which just runs on smartphones or wearables anymore. Mobility is a bigger umbrella which covers a variety of devices with varied functionalities. IoT has paved the way for a variety of form factors in mobility. As more and more devices get connected to the internet, there are more data collected and can be processed.
Examples of simple machine learning solutions –
1) A simple camera on the traffic signals that can monitor and streamline the traffic flow
2) High performing vegetable classier that classifies good and rotten veggies before packing and shipping
3) Real-time translation using an audio headset like Google buds
Why is machine learning important? What value does it add to the existing solution?
ML solutions save lots of time and money by letting users finish their tasks quickly and efficiently. Human time and efforts can be utilized where real human intelligence is needed to solve a problem. Letting machines understand and classify the things on its own in detecting new items by comparing against the old data it may have helped in simplification of the user interaction for the customer and improving productivity for the business.
What are the different machine learning solutions available for mobility solutions? Is it easy for mobile developers to use ML in their solutions?
High network availability and high processing powers of the mobile devices are providing a way for good ML solutions. This paves way for 3 different approaches – (1) Cloud AI API’s and (2) On device/Federated learning and (3) Hybrid of 1 & 2
1) Utilizing the on cloud AI/ML Api’s – All the 3 major cloud providers (Google, AWS, and Microsoft) provide AI services through API’s.
A) Vision AI
B) Video AI
C) Translate API
D) Natural Language API
E) Speech-to-Text API
F) Text-to-Speech API
G) Recommendations AI
2) On device – mobile machine learning
Running machine learning on one or a few centralized clouds may not provide a smooth experience on mobile. The network latency, data privacy, etc may add complications. Performing machine learning on a device is emerging because of the hardware capabilities and more sophisticated modern tools like TensorFlow Lite for mobile and IoT devices.
TensorFlow Lite –
A) Object detection in image and video
B) Image classification
C) Smart replies
3) Hybrid ML using cloud and on-device learning –
The data captured from the device is not uploaded to the cloud for training the models in the cloud. Rather, the model lives on the device and utilize the on-device processing capabilities to train the data locally and then upload the model to the cloud. Later the model can be distributed to other mobile devices and it is a recursive process. Many modern tools like Firebase ML Kit lets you train the model with a basic understanding of ML. The trained models can later be used on the device to perform ML.
Things to consider in improving the customer experience through ML –
1) Understanding the processing power of the mobility device – Not everything can be processed on the device. Be cautious about the amount of data and the number of CPU cycles the model utilizes. Else, it may be an overkill on the device processing power.
2) Optimal usage of the network – Cloud ML is good. However, having everything on the cloud creates a lot of network dependency and brings latency in application response time on poor network connections.
3) Not violating the data privacy standards – By training our models on the device, we do not send any PII information to the cloud. Only the ML models are sent to the cloud. This eliminates the risk of handling personal data as per CCPA and GDPR mandates.
Share your thoughts on the approach for implementing ML based solutions for mobility applications.