MARLBOROUGH, Mass. – Quantiphi, an applied AI and data science company, has been selected by TensorFlow as an AI Service Partner to enable businesses across industries to benefit from AI-based systems and help them innovate faster, solve smarter, and scale bigger. Quantiphi’s knowledge and experience in building and implementing enterprise machine learning (ML) solutions for diverse applications such as medical imaging, video analytics, and natural language understanding help enterprises looking to accelerate their business goals with ML and AI.
“We are delighted to be named as a TensorFlow AI Service Partner. The ecosystem provides us with tools that help our ML researchers and engineers across the entire lifecycle of development, deployment and management of AI applications,” said Asif Hasan, Co-founder, Quantiphi. “Continuous innovation in the TensorFlow ecosystem with tools like the TF Data API, TF Serving, and the wider TFX platform allows us to architect systems that can push the state-of-the-art and drive increased adoption of machine learning in the enterprise.”
As one of the early adopters of TensorFlow, Quantiphi has integrated the ML models into customers’ end-to-end workflows to transform and scale operations. Earlier last year, Quantiphi was named a TensorFlow Trusted Partner for Machine Learning-led Business Transformation. Being a TensorFlow AI service partner further reinforces Quantiphi’s ability in solving what matters.
“Quantiphi continues to be a valued partner in implementing Tensorflow in production, at enterprise scale. We are thrilled to continue our collaboration and see the diverse AI solutions architected by Quantiphi for its customers” said Amy Hsueh, Tensorflow partnerships lead.
Quantiphi has demonstrated expertise in TensorFlow and ML solutions since early 2016. Quantiphi collaborated with Google to implement distributed training on Cloud AI platforms to achieve higher speeds by using a hybrid distribution strategy. Quantiphi also implemented GPU-based RNN cells and an all-reduce strategy for gradient aggregation with speed on GPUs. This reduced the overall training time from 21 days (approx) to just a few days. These strategies provided support to multiple user segments and improved the performance of the models.
Quantiphi recently achieved great success in using TF for training a DistillBERT model on custom datasets. The trained model powered a Question and Answering system for mining insights from pharmaceutical documents. By using Tensorflow Serving for deployment, Quantiphi attained a sub-second inference time that was necessary for ensuring an optimal user experience.
Quantiphi also succeeded in using TensorFlow for writing custom VNet-style architecture for the development of 3D image segmentation systems that can capture lung CT scans and accurately identify abnormalities. TF Estimators and TPUEstimators (and other TPU supported ops) utilized to train models on multiple GPUs and TPUs respectively helped Quantiphi to scale the model training process and expand the benefits of the technology for its customers.
Quantiphi leveraged TF Serving to deploy the models on Kubernetes clusters and improve the efficiency of the deployment process. This showcases Quantiphi’s diverse capabilities to create real-world impact by extending the benefits of TensorFlow to enterprises and providing them the best solutions.
Quantiphi is an award-winning applied AI and data science software and services company driven by the desire to solve transformational problems at the heart of business. Quantiphi solves the toughest and complex business problems by combining deep industry experience, disciplined cloud and data engineering practices, and cutting-edge artificial intelligence research to achieve quantifiable business impact at unprecedented speed. We are passionate about our customers and obsessed with problem-solving to make products smarter, customer experiences frictionless, processes autonomous and businesses safer by detecting risks, threats and anomalies. For more on Quantiphi’s capabilities, visit www.quantiphi.com