You are here

Lenovo tools boost AI deployment in businesses

Lenovo intelligent Computing Orchestration (LiCO) is designed to overcome pain points for enterprise customers, academia and research institutions


ARTIFICIAL Intelligence (AI) initiatives are at the front and centre of many enterprises and research organisations - and one man has had the privilege of being in the thick of the action when it comes to customer engagements around the globe.

He is Bhushan Desam, global business leader for AI in the data centre group at Lenovo.

Dr Desam said: "Although customers are seeing early successes in prototyping efforts, they face a clear pattern of challenges when it comes to implementing AI projects to meet their end goals."

Here are a few examples of AI challenges that customers have shared with him from their AI journey:

Market voices on:

  • A well-reputed finance firm based in New York City is facing challenges to fulfil the needs of various development teams demanding their choice of frameworks and versions.
  • A renowned healthcare organisation in the US is struggling to manage multiple AI projects in a shared environment of cluster resources.
  • A manufacturing customer in Japan is finding it challenging to reduce the time taken to train deep learning models.

A common challenge among many customers across numerous industries is the lack of highly-experienced data scientists or AI engineers, said Dr Desam.

"Not only does the lack of qualified personnel make these resources very valuable (and expensive to hire), there is also a need for tools to make them as productive as possible."

He noted that time and again, customers seem to be coming across the same few challenges.

"So if the challenges are common among diverse industries, how do you get past them to implement AI projects successfully?"

One distinct aspect of AI compared with other enterprise applications is that it is mostly driven by open source, said Dr Desam.

Thus, there are almost no prepackaged AI offerings where information technology (IT) has experience in either managing effectively or scaling as demand arises, he said.

"This is where Lenovo steps up with their experience and leadership in high performance computing (HPC) to create suitable AI solutions for both enterprise and HPC customers."

The Lenovo intelligent Computing Orchestration (LiCO), among others, was designed to overcome these pain points for not only enterprise customers implementing AI, but also those in other multi-user environments such as academia or research institutions using clusters for both HPC workflows and AI development, Dr Desam shared.

LiCO simplifies resource management and makes launching AI training jobs in clusters easy. It currently supports multiple AI frameworks, including TensorFlow, Caffe, Intel Caffe and MXNet. Additionally, multiple versions of those AI frameworks can easily be maintained and managed using Singularity containers.

"This consequently provides agility for IT managers to support development efforts for multiple users and applications simultaneously."

Dr Desam said that as AI activity ramps up and comes under IT management, many organisations are considering scale-out deployments that offer flexibility and best total cost of ownership (TCO).

T[/ ]o accomplish this, organisations need complementing tools to manage AI development in scale-out environments, he said.

"This is where LiCO shines, giving the flexibility to run . . . on the customer's hardware platform of choice and workloads with a potential for reducing TCO up to 35 per cent compared with purpose-built scale-up systems."

Through distributed training management capabilities, it is easier to launch a parallelised training task in LiCO on multiple servers, he said.

This advantage reduces time-to-train requirements, and allows for faster experimentation, by efficiently scaling computing resources in clusters.

Dr Desam added that LiCO features transfer learning using pretrained models, which reduces computing demands compared with model training from scratch.

"Data scientists and AI engineers require tools to simplify their workflows in repetitive tasks, such as hyperparameter tuning or model set up based on well-known network topologies."

He added: "Command-line monitoring of training jobs can be both time-consuming and error-prone. LiCO addresses some of these critical needs by incorporating intuitive GUI (graphical user interface) driven workflows and monitoring tools that can reduce time for setting up models by up to 25-50 per cent."

Researchers at North Carolina State University (NCSU) are reportedly using LiCO to develop AI models for analysing large amounts of geospatial image data to map various features.

Raju Vatsavai, the principal investigator of this effort, said: "When it comes to using advanced AI solutions such as deep learning, application scientists are bogged down by two challenges: which framework to use; and how to set up the training (parameter tuning and computing infrastructure) environment.

"These two impediments to wider adaptability are effectively reduced by Lenovo through their LiCO framework. As a result, scientists from diverse application domains can now focus on extracting valuable insights in a timely manner from their big data and not waste time on putting together mired open source solutions to solve their problems - or worse yet, not using the best solution at all."

Dr Desam added that Lenovo is working with a number of partners on AI initiatives that will provide "real world solutions" for its customers.

"By combining a diverse set of data science expertise with deep industry knowledge of industry segments including health care, life science, energy and manufacturing, we're able to provide solutions that meet customers where they are."

He said that collaboration with partners such as Intel, Nvidia, Mark III systems, and byteLAKE "greatly expands" the resources and expertise that Lenovo is able to provide.

"By relying on vast resource and expertise from our partners and Lenovo's research & technology teams, Lenovo has built AI proof-of-concepts that detect tumours in human livers to aid healthcare providers and improve quality control for manufacturers using image recognition models."

Carlos Morales, head of deep learning systems for AI products at Intel, said: "Intel strongly believes that the potential of AI across industries will be realised by democratising the technology and easing its implementation.

"The combination of expertise from Intel and Lenovo provides an easy path for customers to get started, resulting in a greater time-to-value and ROI (return on investment) as AI projects come to fruition."

  • A version of this article was first published in Xperience, Lenovo's data centre newsroom, on March 21, 2018