All sessions of Transformation 2021 are now available on demand. Now look.

MLOPS sits at the intersection of machine learning and information technology operations operations, developer operations operations (devOps), data engineering and machine learning. MLOPS aims to bring machine learning algorithms into production.

While similar to DevOps, MLOPS is based on different roles and groups of skills: data scientists who specialize in algorithms, math, simulation and developer tools, and administrators who focus on upgrades, product deployment, resources and data management and security. Focuses. While there is significant business value for MLOPS, implementation can be difficult in the absence of a strong data strategy. Enterprise MLOP Kenny Daniel, founder and CTO of Algorithmia, the company behind the platform, spoke with VentureBeat about MLOPs, its benefits and its challenges.

This interview has been edited for clarification and promotion.

VentureBet: How does MLOPS work?

Kenny Daniels: MLOPs are using DevOps and software software engineering best practice lessons in the world of machine learning. MLOPs include all the capabilities that data science, product teams and IT operations need to deploy, operate, manage and secure other potential models in machinery learning and production. MLOPS combines the practice of ALO / ML with the principles of DEOPS to define the ML life cycle that exists alongside the Software Development Lifecycle (SDLC) for more efficient workflows and more effective results. It aims to support continuous integration, development and delivery in production on the basis of AI / ML models.

We break down MLOPS into 10 main capabilities in the three-step ML lifecycle (development, deployment, rations operation) deployment and stages operation phase. Around our ML lifecycle deployment phase:

  1. Training Integration – Extensive language and structural support for any DS tooling.
  2. Data Services – Native data connectors for popular platforms, as well as permission and access controls.
  3. Model registration is integrated with your docs, IDE and SCM, search and tagging so you know the origin of all your models in production.
  4. The service and piping of the algorithm – approval of complex assemblies of models required to support the application – should be hand-maintained.
  5. Model Management – How you can build version management, A / B testing, resource and licensing control and history management.

Around the operational phase, there are also five key capabilities:

  1. Model operation operations – how you control usage and performance in a product involves the approval process and permission control.
  2. Infrastructure management, including fully automated infrastructure, redundancy, osc tosk aling gender, -n-premise, cloud and multi-region support.
  3. Monitoring and reporting – MLOPS’s “who, what, where, why, why and when” visibility.
  4. Rule for internal and external compliance, ling ging, reporting, customer metrics.
  5. At all stages including security, data encryption, network security, SSO and proxy compliance, permissions and controls.

VentureBet: The nature of AI deployment depends on the maturity of the organization. In this situation, where does the organization need to be to prepare for MLOPS?

Daniel: MLOPs become relevant when trying to bring machine learning models into production. This will usually happen only after the establishment of the Data Science program and the projects are going well. But it is too late to wait until the model is built and if the MLOPS story is not resolved, it will be too late to build.

VentureBeat: What are the common mistakes in MLOPs?

Daniel: Leaving the responsibility to personal data scientists to navigate IT / DevOps / Security departments on their own. This sets a recipe for failure, where success depends on a specialized team navigating a completely different software engineering domain. We’ve seen a lot of companies that will take teams of data scientists and machine learning engineers and set them up with loose building models. Where they have created a model and need to prepare it to deploy and handle product traffic, there are many things that need to be in place. These are things that are considered mandatory in the modern IT environment, not just for machine learning: source code management, testing, continuous integration and delivery, monitoring, alerting and managing the software development lifecycle. Being able to effectively manage multiple services and many versions of those services is especially crucial in machine learning, where models can be retrained and updated on a continuous basis. That’s why for companies, “What is our MLOP story?” It is crucial to answer the question of no. And what is the organization process from data to modeling going into production?

VentureBeat: MLOPS What is the most common use case in?

Daniel: Large businesses use us for mission-critical application. The most common usage cases we see are those that are important for agility, accuracy or scaling of complex applications to gain momentum in the market; Any place where content has an impact on the cost of a fast transaction. For example, Merck accelerates the discovery of drugs and the analysis of complex compounds for vaccine development. EY speeds up the investigation of fraud by frequently updating better models and reducing false positives by 30% with better performing models. Raytheon will support the development of the US Army’s Tactical Intelligence Target Access Access Node program.

VentureBet: How did the advent of low-code / no-code help / hinder MLOOPS?

Daniel: I’m generally skeptical about less / no code solutions. The good thing is that because they usually give opinions about the apps they create, they often come up with a solid MLOP story out of the box. The downside is that while they may be quick to work on a simple demo, most real-world applications will have complexity that goes beyond what no-code tools can support. Customization becomes important for application in production.

VentureBeat: Developers quickly moved to Devsek ecpas as developers realized we would also integrate security operations into development. Is there a security element for MLOPS?

In our research, governance as well as safety is the top challenge that organizations face when configuring MLM models in production. There is a fairly security element to MLOPS, and it is converging with more traditional data and network security. Enterprise-grade security is definitely something ML engineers should consider as the first-order capability of any MLOPS domain. I’m talking about data encryption at rest and in flight, unique model control, API pairing, private and public certification authority, proxy support, SSO integration, key management, and potential air-gap deployment support for high security usage.


VentureBet’s mission is to become a digital town square for technical decision makers to gain knowledge about changing technology and transactions. Our site provides essential information on data technology and strategies to guide you as you lead your organizations. We invite you to become a member of our community for:

  • Up-to-date information on topics of interest to you
  • Our newsletters
  • Gated idea-leader content and discounted access to our precious events, e.g. Transformation 2021: Learn more
  • Networking features and more

Become a member