The platform enables several use-cases including news feed personalization, language-modeling and real-time unsupervised learning in addition to allowing data science teams to plug and play their own models
Abacus.AI is the world’s first enterprise scale end-to-end real-time MLOps and DLOps platform. Customers can stream real-time events, such as clickstream data, social media interactions, online purchases, media views and readings from iOT sensors to Abacus.AI through a streaming API. Abacus.AI will process and transform that data, train deep learning models, and generate contextual predictions in real-time.
Real-time deep learning systems are very powerful and form the core AI models of social media platforms such as Facebook, search engines such as Google, video platforms such as YouTube and mobile apps such as Uber. These systems enable virtuous feedback loops that increase customer engagement and retention. For example, on YouTube, these models recommend live sports videos when a user has just watched an Olympics video and late night comedy clips when their intent changes and they are in the mood to laugh.
With Abacus.AI, all organizations, not just big tech, can build large-scale, real-time enterprise AI systems with ease. Data science teams in these organizations can either use Abacus.AI’s neural architecture search to train models, or specify their models on popular frameworks such as TensorFlow and Pytorch, and let Abacus’s end-to-end AI platform take care of the rest.
Abacus.AI features all the key components of an end-to-end AI service, including easy set up of data pipelines, data cleaning and transformation, model training and hosting, a real-time ML feature store service, model monitoring, and explainability. Abacus.AI follows the principle of keeping simple things simple, while making complex things possible. The service allows ML beginners to build and host simple ML models through an intuitive UI, while advanced ML practitioners and experienced data scientists can leverage APIs, and write python code and SQL queries to build their own models.
The real-time deep-learning system can be used for both supervised and unsupervised learning and can be applied across a variety of use-cases including newsfeed personalization, personalized search, natural language processing, e-commerce recommendations, predictive maintenance in factories, cloud spend monitoring, and alerting systems that detect unexpected events.
While data science teams have the option to bring their own models, customers can also choose to use Abacus’s expert AI engine to develop custom deep-learning models that are based on the use-case and dataset.
These deep-learning models choose the best neural architecture from an array of different types of NNs including LSTMs, RNNs, Transformers and Variational Auto-encoders. In order to train these custom models, Abacus.AI has invented several new neural architecture techniques.
These techniques have been published as research papers at top AI/ML conferences including NeurIPs and ICML. A list of recent publications can be found here – https://abacus.ai/publications. Recently, the company applied its techniques and participated in the prestigious CVPR competition, Unseen Data in Neural Architecture Search, and placed 2nd.
Once trained, these models can easily be evaluated using a comprehensive model metrics dashboard, pushed to production and continuously re-trained on a regular schedule or whenever significant feature drift is detected. Abacus.AI provides dashboards to monitor latency, traffic, errors, and feature drift of models in production. This removes all the heavy lifting needed by data science teams to operationalize and run models.
To-date, over 6,000 customers have used Abacus.AI to train over 20,000 models and several of them, including 1-800-Flowers, Flex, Recorded Books, Daily Look, and Prodege use Abacus.AI in production for several of their AI use-cases.
For more such updates and perspectives around Digital Innovation, IoT, Data Infrastructure, AI & Cybsercurity, go to AI-Techpark.com.