Google unveils tools for developers to build machine learning and AI into their products

Google LLC deploys a number of tools for developers to incorporate machine learning and artificial intelligence into their applications using high-performance AI models and solutions.

Today at Google I/O, the company’s annual developer conference, they announced a number of new tools for TensorFlow. TensorFlow is a free, open-source software library for machine learning and AI, with a special focus on neural network training and inference across a variety of architectures, from server to mobile.

Google is increasing support for AI models such as generative AI and image diffusion models. TensorFlow For developers, it makes it easier to use the library and integrate it into their applications. Generative AI has become very popular recently with the introduction of OpenAI LP’s chatbot ChatGPT, which can conduct human-like conversations, and the art-generative AI Stable Diffusion, which can create beautiful and surreal artwork.

Keras, a high-level Python library for interacting with TensorFlow, gets two updates aimed at making it easier for developers to add AI capabilities to their apps with just a few lines of code. The first is KerasCV for computer vision and the second is KerasNLP for natural language processing.

Whether developers want to invoke text generation AI or image generation AI, KerasCV or KerasNLP can be used to provide prompts and receive output directly within the app with just a few lines of code. . Since these new additions are part of Keras, they have full access to the TensorFlow ecosystem.

Google also updated DTensoris a specialized tool for training large AI models that enables parallel scaling. As AI models grow larger, they become more difficult to train because they cannot be trained on a single device. Traditionally, developers have had to split or shard models across multiple processors, such as graphics processing units and tensor processing units.

With this update, DTensor enables larger, more performant training and fine-tuning, bringing it on par with industry benchmarks for training large datasets. As a result, the developer can be confident that his AI model will be ready faster and more efficiently.

Since much of the work in machine learning begins with research, Google has also made it easier for researchers to develop their models by moving them to TensorFlow. Jaxis a powerful framework for converting numerical functions to TensorFlow using an application programming interface called JAX2TF. This means researchers can keep developing new models, and when they’re ready to go into production, they just pipe it through the API and they’re ready to go.

Google also has a machine learning and AI solution building space called ML Hub. The hub will allow developers, engineers and stakeholders to define what they want to do and use cases, and Google will provide them with education, templates, modules and tools to build bespoke AI solutions from our ecosystem. To do.

Google has a number of different tools for bringing machine learning and AI into developer apps, but they are so complex and distributed that developers have to rely on each other to achieve specific desired results. Finding out what you want can be difficult.

MediaPipe makes it easy to deploy machine learning on mobile

Not all AI takes place in giant server farms. Some models are small enough to run on more constrained computing devices such as mobile phones, and Google upgraded them to make that easier. media pipe.

MediaPipe makes it easy to build, customize, and deploy on-device machine learning solutions for portable edge-based computing that run on mobile devices, desktops, and the web. With on-device capabilities, machine learning can perform gesture detection, such as hand and face movement monitoring, enabling powerful device capabilities. It can also be used for many other features such as automatic translation, background blurring, etc.

One particular use case for MediaPipe and small AI models is how it can be used for accessibility, especially for individuals who cannot use their hands or feet to access their device. For that reason, Google developed “project game faceis a computer-controlled interface that uses facial expressions that can control mouse movements in video games to assist gamers with disabilities.

Google partnered with Lance Carr, a gamer with a rare form of muscular dystrophy. His house burned to the ground and the equipment he usually used to play games such as World of Warcraft was destroyed. Google engineers set out to use MediaPipe to allow webcams to control the gaming experience. For example, you can raise your eyebrows and click and drag your mouth, or move your lips to one side to move the cursor.

All of this could be done on a single machine without needing anything overly powerful, allowing Carr to get back into the game and fly across Azeroth again.

Project Gameface is just one of many potential possibilities for portable AI, but it is very powerful. “Controlling a computer with a funny face? That’s pretty cool,” Kerr said.

Image: Pixabay