End to end AI platforms can not only provide businesses with a unified AI strategy but also provide integrated tools for managing data annotation projects of any size. In this talk of Computer Vision DevCon 2020, Matthew Zeiler, Founder and CEO of Clarifai, an independent artificial intelligence (AI) company, talked about common challenges companies face while deploying AI, and how Clarifai’s complete AI ecosystem can help companies achieve their AI goals.
As a founder of Clarifai, Zeiler works on simplifying the complex challenges related to image and video recognition and making it accessible to all. With Zeiler’s tremendous experience in the field, he has built Clarifai’s problem-solving AI ecosystem which has been explained further in his talk.
Clarifai’s AI Ecosystem
To start the talk, Matthew explained three common challenges which restrain companies from fully deploying artificial intelligence in their companies — siloed teams, the requirement of more profits, and lack of expertise. According to Zeiler, for a large organisation, the majority of their AI efforts are happening in different teams at different locations, and not using the same tools. In such scenarios, adopting a platform like Clarifai can really unify all those AI efforts, which not only allow business leaders to monitor the progress of these AI efforts but also build a centre of excellence around the tool.
Further Zeiler explained the requirement for more profit, where companies need to be innovative and have differentiation of propositions in order to drive more profit. And the speaker believes that “AI can become an accelerant for that.”
Thirdly, he explained how the lack of expertise could majorly restrain companies in fully leveraging the benefits of AI. This brings down the commoners to be unaware of where to start with and how to build that position. “You might not have the talent in house, just Google, Microsoft and Amazon, which can be a big challenge in running the AI efforts. And this is where adopting an end-to-end platform like Clarifai makes it easy to get AI into your organisation.”
A system like that easily allows organisations to build a centre of excellence around the tool which easily syncs with your existing infrastructure and IT department. And that’s where Clarifai’s ecosystem comes in. It starts with the end-to-end platform for the AI lifecycle, which consists of all relevant tools businesses need to build a world-class AI. “Our portal is kind of your entry into AI,” said Zeiler.
To use the platform Clarifai, one doesn’t have to be a developer neither have to be a PhD in artificial intelligence. All the platform requires is the user to log in and within minutes to be up and running. Also, the platform comes with APIs that will let the user connect as a developer into the platform, which will enable them to send data to train the models, do searches and run at production scale.
The company has deployments with all of its three competitors, AWS, Azure and GCP, and is able to run on the cloud as well as servers. Additionally, Clarifai can offer data labelling and model building and expert services to get the business up and running with expert tools in no time.
AI Lifecycle
Explaining further, Zeiler showcases Clarifiai’s end-to-end AI lifecycle, where all starts with data that we support today, such as images, videos as well as texts. To facilitate the process, firstly, the data needs to label, which is the annotation process. With the built-in AI, it can help businesses automate the labelling process, which allows in scaling the annotation process really large. “And all of that annotated data is indexed in a large, distributed database so that business leaders can search over it. And this is important as a product itself,” said Zeiler.
The model supports searches by human applications, which could be either dog or cat, which becomes an essential aspect of the training process. Further, this can be used to find the subset of data that one is willing to train their model on. All these data are then combined with algorithms to stay up to the speed with all the developments in the research community. While running the model, the users get the choice of running fully hosted in either cloud, on-premise or mobile and IoT devices, where even the deployment phase has been fully automated.
To explain it, Zeiler said, “We allow you to take a collection of that data, we actually call it collectors to syphon off data from that model running in production to automatically annotate it or send it to a human for review. So if it’s really confident that there’s a dog in this picture, you can probably trust that label and treat it in the training phase of the next version of the model. It is not confident that there’s a dog, maybe you want to send it to a human to review if there’s a dog or not. That’s the idea of active learning built into our platform.”
Clarifai API: Advanced Developers Tools
- API clients in your language of choice to fully integrate with our platform.
- End-to-end APIs to process inference on image, video and text inputs.
- Industry-standard JSON HTTP API structure to integrate AI into your business outcome.
- Complete API defined by photo-buffers with gRPC API clients for high-performance communication.
Wrapping Up
Further, the speaker showcased some examples of the model like a general model for social media consumers; colour model to identify dominant colours; demographics models to predict age gender as well as culture; face detection model to detect human faces; object detection model for the apparel industry, to name a few. These varieties of use cases highlight that the model can be trained and customised for several sorts of applications — starting from eCommerce and travel companies to departments of defence.
