AI with Purpose-Built Deep Learning Solutions-as-a-Service

AI and deep learning are identifying new opportunities, revenue, and efficiencies in multiple industries.



The 42nd IT Press Tour had the opportunity to meet with Rodrigo Liang CEO and co-founder, Marshall Choy SVP Product, Prabhdeep Singh VP Software Solutions, Arvind Sujeeth VP Software (Compilers), Amy Love CMO, and Keith Parker Director of Content and Ecosystem Marketing at SambaNova Systems.


SambaNova equips customers to use AI and transform their business for the AI-enabled world in weeks rather than months or years. They deliver Dataflow-as-a-Service, a complete AI solution including next-generation hardware, software, and pre-trained models that overcome the limitations of legacy technology to power the largest AI models.


Background


The market is undergoing another disruptive and fundamental shift with AI taking center stage.


Companies that have already joined the AI revolution are outpacing the market and using AI to discover new opportunities, unlock new revenue, and boost efficiency. The next wave of innovation in this space has arrived and is being driven by deep learning with significant advancements in computer vision, large language models, and recommendation algorithms.


To take advantage of these cutting-edge deep learning capabilities, organizations have needed to invest significant effort and expertise to develop and deploy new models and have needed to upgrade from their legacy AI infrastructures, which are often unable to manage even 1% of a state-of-the-art deep learning model.


SambaNova helps customers overcome these limitations with Dataflow-as-a-Service, the leading AI platform which brings together the best hardware, software, and pre-trained deep learning models. This reduces the time required to deploy these cutting-edge AI capabilities from years to weeks and helps these organizations move fast with an innovative portfolio that delivers value today and is designed as a foundation for the future


Organizations across banking and insurance, manufacturing, healthcare, energy are leveraging state-of-the-art computer vision, large language models, and recommendation systems to improve their business and achieve substantial revenue gains and operational efficiency.


The Evolution and Adoption of AI


Over the past two decades, AI has undergone a generational shift from developer-driven scale-out computing, to consumer-driven embedded devices, to data-driven dataflow computing.


Successful companies today, and in the future, are using AI to improve CX, UX, and business performance.

  • According to IDC, by 2025, every connected person will have a digital interaction every 18 seconds.

  • According to PWC's Annual Global CEO Survey, 63% of CEOs believe AI will have a larger impact than the internet.

  • According to McKinsey's The Executives' AI Playbook, $15 trillion will be added to the global economy in the next decade by AI.

AI is driving massive industry shifts, accelerated adoption and innovation of technology, and rapid adoption and deployment of AI at scale.


According to SambaNova's commissioned research:

  • 70% of companies surveyed plan to allocate more than $100 million of IT budget toward strategic technology goals.

  • Drivers of AI/ML investment are innovation (40%), operational efficiency (25%), competition (22%).

  • 75% say improving access to deep learning is very important for fostering competition and innovation.

  • The biggest AI/ML challenges at difficulty customizing models (50%), restrictive computing architectures (35%), and lack of trained talent (28%).

Deep Learning Era


We have moved from AI 1.0 -- digital transformation and machine learning to AI 2.0 with the industrialization of computer vision, highly flexible, general-purpose models which enable the human-level performance of some tasks to be surpassed.


Organizations looking to adopt deep learning are struggling with technical challenges and competing for limited expertise. GPU/CPU-based AI infrastructure is outdated. There is tremendous competition for AI talent. Deep learning advancements are outpacing the ability to deploy them.


This is resulting in a deep learning deployment gap. It takes 18 months to hire a team, build infrastructure, train and deploy a model. Model sizes increase 10X every 12 months. Compute requirements increase 2X every 3.4 months.


SambaNova addresses these challenges with advanced deep learning capabilities, which results in faster time to value and reduced complexity. The deep learning platform is purpose-built for innovation and has the ability to handle rapid AI innovations that require a 10X increase in compute requirements and model size every year.


Dataflow-as-a-Service is an integrated hardware/software platform with deep learning models-as-a-service (computer vision, natural language processing, or recommendation systems) deployed in weeks rather than months. This has attracted the most attention from companies in the energy, financial services, healthcare, manufacturing, and the public sector.


Platform Overview


State-of-the-art AI is the province of the elite few.

"In the weeks since its arrival, GPT-3 has spawned dozens of other experiments that raise the eyebrows in much the same way it generates tweets, pens poetry, summarizes emails, answers trivia questions, translates languages, and even writes its own computer programs, all with very little prompting." -- New York Times 11/24/2020

The old way of processing data for AI was kernel-by-kernel. Performance was bottlenecked by memory bandwidth and host overhead. Dataflow spatial programming eliminates memory traffic and overhead and increases parallelism.


AI Use Cases


Financial Services:

  • Customer fulfillment and retention - Chatbots for faster MTTR in customer service, customer churn prediction, and prevention.

  • NLP for insurance cost reduction - Can take into account data points from vast amounts of data that humans simply can't.

  • AI-based credit scoring - Increases speed, the accuracy of credit risk models and gives quick access to credit for BNPL.

  • Algorithmic trading - Reduce compliance risks, analyze more signals, aid in alpha generation strategies,

  • Robo-advisers - $8T AUM in wealth management is slowly being moved to Robo-advisers.

  • AI-based fraud detection - 5% of global GDP transactions are money laundering and $80-100B lost in fraud transactions. KYC scenarios.

  • AI-based Regtech - Respond to regulatory requirements quickly and effectively.

Healthcare:

  • Disease detection – Data analysis using NLP for diagnosis, high-res imaging analysis for disease diagnosis in radiology, pathology image analytics.

  • Drug Discovery – ML analysis of patient data, clinical trials, publications for new drug applications. Protein folding prediction.

  • Targeting and detection of evolving pathogens – detect and reduce antibiotic resistance.

  • Real-time patient monitoring.

  • Healthcare supply chain management.

  • Dynamic pricing of health insurance.

  • Supplementing underequipped points of care in developing/underdeveloped/ disaster stuck places.

Manufacturing:

  • Defect detection – High-res imaging to detect a defect on assembly and QA lines.

  • Predictive maintenance – time series models to predict machine failures in advance.

  • Yield optimization/enhancement - RL used for robotic control, extruder operations, and chemical process optimization.

  • Autonomous machine control – various control inputs can be optimized in a closed-loop system for reduced energy, better quality, sustained reactions.

  • Generative design – Use generative models to design better than traditional topology optimization.

  • Inventory management – predictive models for inventory replenishment.