NDR – The AI Conf 2021

What’s your experience working with model serving and model inference? Do you know what are the potential ways to improve for it?
When moving model from training to production. There are things to be concerned, for instances, SLA, model management, compute, data, cost and security. This session Mia is going to share how to optimize the model inference from software, hardware and network perspectives. Tips about model inference optimization.

In many cases, a data scientists just needs a Jupyter notebook on a powerful virtual machine to be productive. However, there are some cases where more powerful cloud services can be helpful. We will give overview of Azure Machine Learning, a platform ML service on Microsoft Azure cloud, and consider some best practices of using it for repeated experimentation, hyperparameter optimization and running parallel sweep jobs. We will focus our specific examples around CORD-19 dataset of scientific papers on COVID-19.

At Accenture Industry X, we understand the importance of including machine learning techniques in the services offered to our clients in order to improve efficiency,help with quality control and predictive maintenance. We aim to show how an alarm status for a motor can be predicted with 10 days in advance , by using as input historical daily sensor readings and data derived from such readings.
By doing so, we can optimize maintenance actions and prevent downtimes.

Most papers and conference talks focus on model architecture and how it improves on the state-of-the-art.
Not this one. In this talk, we focus on system design. How do we design systems for low-latency, high-throughput serving for recommendations and search? In this talk, I’ll share a 2×2 framework of online vs. offline and retrieval vs. ranking. We’ll also go through examples of how companies like Alibaba, Facebook, and DoorDash serve their recommendation and search.

The session focuses on the generative aspects of Natural Language Processing. Barbara will introduce how AI new advancements are being used for text generation. This will be followed by the review of existing systems and available tools that could be used for creating their own solutions.

The audience will get three key takeaways:
– NLP techniques for text generation.
– Tools that can be used to generate text.
– How text generation is and can be used in the real world applications.

Machine translation has come a long way over the past 70+ years.
Today’s state-of-the-art Neural Machine Translation (NMT) has created a paradigm shift. This radically different approach to the challenge of language translation uses deep neural networks and artificial intelligence to train models for different language combinations and content types.
In this talk we’ll cover challenges and solutions on building and improving upon these advances and turning them into powerful real-world products.

Using machine learning techniques such as neural networks and evolutionary computation, it is possible to build models to predict not only how the pandemic spreads, but also how it could be mitigated. I will review Cognizant’s Evolutionary Surrogate-Assisted Prescription (ESP) technology, its application to this task, the XPRIZE competition that Cognizant sponsored based on it, and lessons learned from the competition, as well as opportunities for using this technology on other decision-making tasks in healthcare, design, and business. – AI support for decision making requires not only predictive models, but prescriptive models as well – ESP is a powerful new technology for discovering prescriptive models – Evolutionary AI can be used to bring together insights from a community of human experts – AI can play a significant role in coping with COVID-19 and future similar challenges