Tag Archives: ML

Posted on Author Ramkumar Balasubramanian Categories Cloud

AI adoption increases with Cloud Machine Learning as a service

The basic ingredient required for AI is a successful machine learning model. However, to create and run the model it is important to have the right capacity of infrastructure, followed by good domain knowledge and a large amount of data. A machine learning model is defined as a software entity created with algorithms and training data. The success of the model depends on getting the right training data with precisely tweaked algorithms. Running the model requires huge infrastructure with GPU / TPU CPU capacity and this makes most of the enterprises wary of investing in AI. The capex model for AI infrastructure calls for a very large investment, plus with the technology developing at a fast pace, enterprises will find it difficult to keep changing the infrastructure.

Essentially for Businesses to take advantage of AI, they must have a huge capacity investment. However, cloud comes to the rescue by offering infrastructure capacity with a lesser initial investment and pay as you use models. In the following blog let us look at the models offered by various cloud service providers for running machine learning as a service (MLaaS). We look at four major cloud service providers namely Amazon, Microsoft Azure, Google Cloud (GCP) and IBM Watson.


Amazon Machine Learning

SageMaker is an Amazon machine learning framework with built-in models with algorithms for classification, regression, multi-class classification, k-means clustering and so on. SageMaker helps to create the models quickly with advanced algorithms. Apart from this, Amazon also offers huge infrastructure on-demand as well as serverless-processing enabling the models to be run on the most optimized infrastructure. Amazon SageMaker also gives hook to Google Tools like Tensorflow, open-source Keras, Facebook Pytorch etc. A complete MLops (Equivalent of DevOps for Machine learning code) is offered by Amazon.

Azure Machine Learning Platform

Services from Azure Machine learning can be elaborated two-fold, Azure Machine Learning studio and Bot service. The graphical drag and drop machine language workflow creation ability is created by Azure using Azure ML Studio. This includes data-exploration, pre-processing, choosing methods, and validating modelling results.The main benefit of using Azure is the variety of algorithms available to play with. The Studio supports around 100 methods that address classification(binary multiclass), anomaly detection, regression, recommendation, and text analysis. It is worth mentioning that the platform has one clustering algorithm (K-means).

Azure serves different kinds of customers. Namely data Scientists, Data engineers, data analysts and so on. Azure’s approach is to provide an end-to-end platform for all types of customers and the product includes model management tools, python packages and workbench tools.

Google Machine Learning Services

Google being an AI-first company offers a variety of AI tools for the developers, enterprise operations, data scientists etc. Google recently started AutoML which requires no programming to develop a Machine Learning model. Google has been a great contributor to open-sources. Most recently they introduced Google BERT, TensorFlow, AutoML etc. Among all the service providers Google has done the maximum contribution to open-source and this, in turn, has improved the adoption of Google Tools. Today TensorFlow is the most widely used development tool amongst the developers. It has different libraries available from multiple open sources making it one of the more popular developing applications on the cloud. Google Cloud also offers a high-end computing environment with TPU processors along with robust data security making it one of the most versatile platforms for development and deployment. Many cloud-native deployments are possible in Google Cloud Platform.

IBM Watson

IBM Watson one of the earliest and very widely used machine learning platforms, has been in existence for some time. It offers a set of services for newcomers as well as experienced service providers. Separately, IBM offers deep neural network training workflow with flow editor interface similar to the one used in Azure ML Studio.


Machine Learning Services offered by the cloud providers.

  1. Speech and text service Translation service
  2. Image classification
  3. Text classification
  4. Speech classification
  5. Facial detection
  6. Facial analysis
  7. Celebrity recognition
  8. Written text recognition
  9. Video Analysis etc

In conclusion, many of the cloud service providers have recognized the fact that business transformation can be brought about by the use of AI technology and provide machine learning as a service so that enterprises can use the readily available models. This helps them in the areas of prediction, personalization, natural language processing, optimization, and anomaly detection. Businesses want a competitive edge and AI plays a key role. And to enable AI quickly, cloud is the way to go.

Cloud Architect

0
Posted on Author Kunal Bhatt Categories Digital

Next Generation Cloud Adoption: Distributed Cloud

Cloud Computing is an evolving discipline. Newer innovations in cloud management are coming into fruition as we speak. What started out as a ‘High-Availability Storage Space’ is now integrated into every function of business. The Cloud opens possibilities for customers to gain benefits and be agile with their workloads. By shifting to cloud they leverage the economics offered by cloud like elasticity, pace-of-innovation, better uptimes and much more, from cloud-based scheduling, cloud-based applications to cloud-based Data-backup and DR. Practically everything has to come prefixed with ‘Cloud-based’ to ensure BAU continues uninterruptedly. However, there is still a pinch of resistance and hesitation seen in organizations when deciding to go for a public cloud model, entirely.

Some prefer private cloud or to an extent are willing to adopt hybrid cloud. Private cloud, is designed in a way that they are, owned and controlled by the customer and operated by the service provider’s teams or the customer’s own technology team and in the hybrid cloud, the public cloud provider manages their set of cloud offerings.

Hybrid Cloud was introduced to further the ‘best of both worlds objective’ for businesses that were not keen on completely abandoning their Legacy Systems in favour of a fully Cloud-based IT Infrastructure. It provided a sort of ‘safety net’ whose requirement was triggered mostly by data security concerns. Distributed Cloud does all this and more.

Distributed Cloud is Cloud-based Technology’s newest offering. Gartner identified Distributed Cloud as one of the top 10 trends of 2020 and the hype around it does not seem to be slowing down and will seemingly continue well into 2021 as well, by the look of things. Distributed Cloud basically leverages Public Cloud to interconnect IT Infrastructure irrespective of Physical/Geographical Location.

Gartner describes Distributed Cloud as “the distribution of public cloud services to different physical locations, while the operation, governance, updates and evolution of the services are the responsibility of the originating public cloud provider.”

Let’s consider the scenario where a business maintains some data on-site, some on private/public cloud and others on edge environments. Maintaining all these complex IT environments require overhead and maintenance to some degree. There is also the issue of all these being physically apart. Not to mention delay/latency concerns. What a Distributed Cloud Arrangement brings to the table is the ability to extend Public Cloud Capabilities to these complex systems and manage all of a business’s spread-out IT Infrastructure.

Cloud-computing involving Distributed Cloud utilizes so-called ‘substations’ as coined by Gartner. These tactically located substations act as a shared cloud pseudo-availability zones with networking, computing and storage capabilities.

Hybrid Digital Infrastructure Management vs Distributed Cloud

In a way Distributed Cloud Management makes up for everything HDIM falls short of. This type of cloud management does not rely on a unified approach to IT Infrastructure Management. It rather focuses on usage-consistency, customization and most importantly governance.

Firstly, Distributed Cloud raises the bar in terms of networking capabilities of IT Infrastructure Clusters. Inter-communication amongst IT clusters whether it is based on-premises and on Public platforms or Edge environments, is a striking feature of Distributed Cloud. This ensures users will have consistency across the board while utilizing the IT Infrastructure. DC also dissipates chances of network failure owing to the presence of sub-stations. This was not possible in a hybrid cloud arrangement.

This uniformity in usage does not hinder customization in Distributed Cloud Systems. Personalization based on the pertinent requirements of a particular location is possible while using distributed cloud. This drives value for the customer as well as the system administrator.

Dev Ops efficiency while deploying high-value services is also augmented by Distributed Cloud. It gives freedom of choice to users when it comes to deciding their preferred cloud clusters/locations. Integrating with Public cloud features allows Distributed Cloud to have the ability to implement innovations like AI/ML based automation capabilities to all IT environments.

Source: O’Reilly- Cloud Adoption in 2020

Another key characteristic of Distributed cloud is its ease-of-governance. If any new policy is introduced at the on-site level, it will be reflected on all cloud-based and edge systems as well. Data security is thus maintainable across the whole IT Infrastructure. This ensures the same level of security at all IT environments regardless of whether it is Cloud-based or on-site. This obliterates the security concerns posed by Hybrid Cloud.

Unifying Public Cloud and IT Infrastructure

To say it in the simplest of terms, Distributed Cloud can bring the unique competencies offered by Public Cloud to IT Infrastructure and make the experience of using cloud-based and non-cloud-based infrastructure less challenging, not to mention the reduction in cost. All this drastically reduces delays to service-delivery and makes the customer-business interaction a delightful encounter.

Source: IDC 2020

But with the unification comes issues like trouble-shooting complexities due to increased chances of interaction between cloud and on-site environments. Replicated data at all these environments also have to be kept track of and secured. So, although it is the same level of security across all platforms, the intricacies regarding the same may increase. Another factor to consider is the cost of deployment. Although operational costs may drop, the resources required to deploy such distributed systems may shoot.

Is this truly ‘The Best of Both Worlds’?

HDIM is constantly described as such but distributed cloud systems may be the new ‘the best of both worlds’ scenario that will see more adoption-rates with businesses requiring more customized offerings that do not compromise on security. But Distributed Cloud is not as ‘tried and tested’ as HDIM and may only look good on paper. that may depreciate ROI, as mentioned earlier. Only time will tell. But once perfected Distributed Cloud Systems are projected to be the future of cloud-based IT Infrastructure management.

Head, Automation Practice

Data References:

https://www.oreilly.com/radar/cloud-adoption-in-2020/

https://www.idc.com/getdoc.jsp?containerId=US46796120/

0