Ever since the COVID-19 outbreak, the disease has claimed the lives of more than 7,000 people, infecting over 183,000, and prompting a wave of travel restrictions, social distancing and enforcement of remote work to contain the disease from widespread.
In an effort to cease the widespread of the disease, companies like Google, Amazon and Twitter, which employ thousands of people, have extended voluntary work from home to all their employees.
Though the nature of few jobs such as data centre manager cannot afford such luxuries, many other jobs that require data-driven solutions seem to have doing well even when working remotely. It is no secret that cloud platforms have made the life of ML-based organisations easy, but the smooth functioning of any ML pipeline is not straightforward.
According to Twitter, the networking giant stores 1.5 petabytes of logical time series data, and handles 25K query requests per minute.
The scale at which these top companies operate requires round the clock monitoring. So, how do these firms cope up in the face of a pandemic?
So far, we haven’t had an issue while using Google or Twitter. What makes these organisations function seamlessly in the absence of their in-house data scientists, ML developers and other data wizards.
Over The Air Model Monitoring
Of all the professions, data scientists, machine learning engineers and other ML-related practitioners do not really struggle when it comes to working remotely. The many dependencies within an ML pipeline require one to store all the components so as to make sure all the features are available both offline and online for deployment.
In fact, most of the ML developers are freelancers and consultants who monitor an ML project by checking the GitHub repository, sharing code via Colab or Jupyter notebook.
Here are a few tasks that are part of any ML developers routine:
- Getting data in and out of the training algorithm and keeping an eye on statistics in the pipeline to make sure that the model in the training environment delivers the same score as the model in serving environment
- Monitoring the model if its quality is degrading with time i.e. ensuring the freshness of a model
- Exporting models
- Launching models
- Monitoring services that host the models to be independently upgraded without breaking their downstream or upstream services.
How In House Tools Help
One thing that separates top companies from other players is their homegrown technologies. Though the fundamental use of these toolkits is almost similar, many organisation prefer customised solutions to their problems. They even make their technology available to others for a fair price. These in-house tools are a great gift to the ML practitioners as they make integrations within the team quite easy.
For example, the machine learning team at LinkedIn built a domain-specific language (DSL) and integrated it with a Jupyter notebook. Most of the model training at Linkedin occurs offline where the teams train and retrain the models every few hours.
Whereas, in the case of Pinterest, a relatively smaller firm, they have their custom-built tools such as the Voyager to assist them in Pinterest’s everyday business. Voyager is used to carrying out tasks like label cleaning, dataset storage, and visualisation.
No matter what the tools are or what the end product is, the notion of remote work for an ML practitioner was made possible due to the vast amount of innovation that occurred in the cloud scene over the past few years.
To get a sense of how cloud facilitates remote work, let’s consider the case of COMPUTE in an ML pipeline, and how Oracle, a major player that offers cloud infrastructure, allows its customers to deploy models without any hassles:
Virtual Machines & Compute
Virtual Machines (VMs) make an ideal choice for running applications that don’t require the performance of a dedicated server. Oracles’ infrastructure is capable of supporting high core counts and high memory bandwidth with which users can build cloud environments with significant performance improvements over other public clouds and on-premise data centres.
Container Engine for Kubernetes
Container Engine for Kubernetes is a fully managed, scalable, and highly available service used to deploy containerised applications to the cloud.
Container Registry allows teams to perform vulnerability analysis, and authentication to control access to a private container image registry. The registry can detect vulnerabilities or security issues before images are ever deployed to containers.
Cloud deployments have an average of 2.3 times lower total cost of ownership compared with on-premise deploymentsreport by Oracle
All top companies offer cloud services at a relatively low-cost today. Besides, the number of options are plenty and diverse.
Keeping In Touch
The job of a data scientist is multifaceted, and one cannot restrict themselves to data collection or curation the whole day. A typical data scientist would have to communicate at least once in a day with his/her team to monitor the results of the machine learning pipeline that has been deployed.
Slack and Google’s Hangouts have been popular with organisations for in-house communication. Now, Google has even rolled out its advanced Hangouts Meet video-conferencing capabilities at no cost to G Suite.
So, it’s essential to make sure that remote working teams have managed devices with the right policies for Wi-Fi, ethernet, and virtual private network (VPN) access, as well as network certificates for a frictionless deployment of above services.
Here are the tools and services that can help any ML-based organisation to operate remotely:
- Google Colab Pro
- Amazon Web Services
- Google Cloud Platform
- Microsoft Azure
- Jupyter Notebooks
- Apache Kafka
- Oracle Cloud Infrastructure
- G Suite
Of the above services, AWS, GCP and Azure have customised AI tools that can handle heavy workloads. The sporadic growth of AI was complemented with the development of tools that offer seamless cross-domain integration and today, most of the ML practitioners are benefiting from it in the face of a global emergency.