Proactive Steps for Securing MLOps Projects // Chris Van Pelt // MLOps Podcast #192 clip

MLOps.community · Intermediate ·🏭 MLOps & LLMOps ·2y ago

Key Takeaways

Chris Van Pelt discusses essential advice for securing MLOps projects, including implementing least privilege, using good security tooling, and tracking vendors from the beginning.

Full Transcript

so my name is Chris Van Pelt I am co-founder and chief information security officer at weights and biases um I like my coffee as a good old French Press maybe with oh look I've got my little Emoji too um french press a little oat milk lately has been more my jam uh sometimes half and half but yeah just a a good french press but it's like good beans you know like the fancy hipster coffee shop for sure what are some things that you can give us as advice as like easy wins to make sure that hey if we are bringing on new tooling we've got our checks and balances and we're good on that and then be just like in general how can we assure success I think first like start thinking about it from the beginning like even when you are a small start up working on something there are you know a few things that will go a long way in ensuring you're going to be able to develop software securely so one is like leas privilege which is it's such It's Tricky um but it's it's really important so like the easiest thing when you're like giving new Engineers uh access to say your AWS account is give them like some permissions that let them do basically like everything that is bad you want to you know limit the permissions of what anyone on the team can do and hopefully get that into terraform and make it really easy to maybe give people temporary uh escalations and then bring them down uh the other pieces are just using good security tooling so uh set up scanning of any Docker containers that being built for for cves if you ever deliver these to Enterprise customers they're going to find them before you do if you aren't doing that which is embarrassing uh and it should be easy to just update uh the dependencies there and have you know a vulnerability kind of management system like dependabot on GitHub will tell you I think it's especially gnarly in the JavaScript ecosystem there's just so many dependencies and so many cves out there but but staying on top of that um is important because the your customers can see it and their teams are going to be concerned about these things yeah uh what else yeah um I mean keep track of vendors like from the beginning when you decide to like use some thirdparty service first step back and say okay what data are we giving them and what classification does it fall into so you can think of like highly confidential um kind of secret but not um terribly confidential and then just low risk and for anything High you should be like confident that they are a mature vendor that has like a sock 2 um type two at the station or some accreditation because eventually you're going to have to tell your customers who your subprocessors are um and they're going to uh you know be pretty concerned if you're using this other you know brand new startup um that maybe won't have those [Music] accreditations [Music]

Original Description

MLOps podcast #192 with Chris Van Pelt, CISO and co-founder of Weights & Biases, Enterprises Using MLOps, the Changing LLM Landscape, MLOps Pipelines sponsored by @WeightsBiases. Chris Van Pelt, the co-founder and Chief Information Security Officer at Weights and Biases discusses essential advice for ensuring success in bringing new tools into your workflow and maintaining security. Chris shares insights on implementing least privilege, leveraging good security tooling, and staying on top of vulnerabilities in your software. Tune in to discover practical tips for developing software securely and maintaining trust with your customers. // Abstract Chris, provides insights into his machine learning (ML) journey, emphasizing the significance of ML evaluation processes and the evolving landscape of MLOps. The conversation covers effective evaluation metrics, demo-driven development nuances, and the complexities of ML Ops pipelines. Chris reflects on his experience with Crowdflower, detailing its transition to Weights and Biases and stressing the early integration of security measures. The discussion extends to the transformative impact of ML on the tech industry, challenges in detecting subtle bugs, and the potential of open-source models and multimodal capabilities. // Bio Chris Van Pelt is a co-founder of Weights & Biases, a developer MLOps platform. In 2009, Chris founded Figure Eight/CrowdFlower. Over the past 12 years, Chris has dedicated his career optimizing ML workflows and teaching ML practitioners, making machine learning more accessible to all. Chris has worked as a studio artist, computer scientist, and web engineer. He studied both art and computer science at Hope College. // MLOps Jobs board https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Website: http://wandb.me/data_talks_club ⁠ --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/sl
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from MLOps.community · MLOps.community · 0 of 60

← Previous Next →
1 Our 1st MLOps Meetup // Luke Marsden // MLOps Meetup #1
Our 1st MLOps Meetup // Luke Marsden // MLOps Meetup #1
MLOps.community
2 Remote Collaboration as a Data Scientist
Remote Collaboration as a Data Scientist
MLOps.community
3 MLOps Manifesto with Luke Marsden from Dotscience
MLOps Manifesto with Luke Marsden from Dotscience
MLOps.community
4 MLOps lifecycle description
MLOps lifecycle description
MLOps.community
5 What Does Best in Class AI/ML Governance Look Like in Fin Services? // Charles Radclyffe // MLOps #2
What Does Best in Class AI/ML Governance Look Like in Fin Services? // Charles Radclyffe // MLOps #2
MLOps.community
6 Life purpose and too many spreadsheets
Life purpose and too many spreadsheets
MLOps.community
7 Explainability, Black boxes and EU white paper on reproducibility
Explainability, Black boxes and EU white paper on reproducibility
MLOps.community
8 Hierarchy of Machine Learning Needs // Phil Winder // MLOps Meetup #3
Hierarchy of Machine Learning Needs // Phil Winder // MLOps Meetup #3
MLOps.community
9 Automatically Retrain Machine Learning Models? Are best practices worth it?
Automatically Retrain Machine Learning Models? Are best practices worth it?
MLOps.community
10 Building an MLOps Team? Key ideas to keep in mind
Building an MLOps Team? Key ideas to keep in mind
MLOps.community
11 Hierarchy of MLOps Needs
Hierarchy of MLOps Needs
MLOps.community
12 Bare necessities for getting an ML model into production
Bare necessities for getting an ML model into production
MLOps.community
13 MLOps and Monitoring
MLOps and Monitoring
MLOps.community
14 How Phil Winder got into Data Science and Software Engineering
How Phil Winder got into Data Science and Software Engineering
MLOps.community
15 Provenance and Reproducibility in Machine Learning; what is it and why you need it?
Provenance and Reproducibility in Machine Learning; what is it and why you need it?
MLOps.community
16 Friction Between Data Scientists and Software Engineers
Friction Between Data Scientists and Software Engineers
MLOps.community
17 MLOps Problems in different size companies
MLOps Problems in different size companies
MLOps.community
18 ML tooling in large companies
ML tooling in large companies
MLOps.community
19 ML Platforms - The build vs buy question
ML Platforms - The build vs buy question
MLOps.community
20 ML Services Gateway at SurveyMonkey
ML Services Gateway at SurveyMonkey
MLOps.community
21 Message buses, Async and sync architecture
Message buses, Async and sync architecture
MLOps.community
22 MLOps #4: Shubhi Jain - Building an ML Platform @SurveyMonkey
MLOps #4: Shubhi Jain - Building an ML Platform @SurveyMonkey
MLOps.community
23 Hybrid Data Science Teams @SurveyMonkey
Hybrid Data Science Teams @SurveyMonkey
MLOps.community
24 How do you handle ML version control at SurveyMonkey
How do you handle ML version control at SurveyMonkey
MLOps.community
25 Doing ML with Personal Information
Doing ML with Personal Information
MLOps.community
26 Evolution of the ML feature store @SurveyMonkey
Evolution of the ML feature store @SurveyMonkey
MLOps.community
27 Developing a Machine Learning Feature Store
Developing a Machine Learning Feature Store
MLOps.community
28 Auto retrain ML models is not the question
Auto retrain ML models is not the question
MLOps.community
29 3 key parts to Machine Learning monitoring
3 key parts to Machine Learning monitoring
MLOps.community
30 MLOps Meetup #6: Mid-Scale Production Feature Engineering with Dr. Venkata Pingali
MLOps Meetup #6: Mid-Scale Production Feature Engineering with Dr. Venkata Pingali
MLOps.community
31 MLOps meetup #5 High Stakes ML: Active Failures, Latent Factors with Flavio Clesio
MLOps meetup #5 High Stakes ML: Active Failures, Latent Factors with Flavio Clesio
MLOps.community
32 MLOps: Airflow Pros and Cons
MLOps: Airflow Pros and Cons
MLOps.community
33 Specific challenges in Machine Learning
Specific challenges in Machine Learning
MLOps.community
34 Current State Of Machine Learning
Current State Of Machine Learning
MLOps.community
35 Humans in the Loop are a defining factor in Machine Learning
Humans in the Loop are a defining factor in Machine Learning
MLOps.community
36 Learning from real life Machine Learning failures
Learning from real life Machine Learning failures
MLOps.community
37 Survivorship Bias in machine learning tutorials
Survivorship Bias in machine learning tutorials
MLOps.community
38 Swiss Cheese model in Machine Learning
Swiss Cheese model in Machine Learning
MLOps.community
39 Resume driven development in Machine learning & software engineering
Resume driven development in Machine learning & software engineering
MLOps.community
40 Who has the highest standards in ML?
Who has the highest standards in ML?
MLOps.community
41 Venkata Pingali of Scribble Data Thoughts on the Current State of Machine Learning
Venkata Pingali of Scribble Data Thoughts on the Current State of Machine Learning
MLOps.community
42 Dependable data and being able to Trust in your Data with Venkata Pengali of Scribble Data
Dependable data and being able to Trust in your Data with Venkata Pengali of Scribble Data
MLOps.community
43 Speed, Trust, Evolution and Scale in MLOps
Speed, Trust, Evolution and Scale in MLOps
MLOps.community
44 More difficult transition for data scientists to become ML engineers
More difficult transition for data scientists to become ML engineers
MLOps.community
45 How many models in prod til I need a dedicated ML platform?
How many models in prod til I need a dedicated ML platform?
MLOps.community
46 Deeper thinking from data scientists around platform blackholes
Deeper thinking from data scientists around platform blackholes
MLOps.community
47 Checkpointing, metadata, and confidence in your data
Checkpointing, metadata, and confidence in your data
MLOps.community
48 Adjacent usecases and multistep feature engineering
Adjacent usecases and multistep feature engineering
MLOps.community
49 Standardization of Machine Learning tools like in Software Engineering with Venkata Pingali
Standardization of Machine Learning tools like in Software Engineering with Venkata Pingali
MLOps.community
50 Reproducability flaws in end to end Machine Learning debugging
Reproducability flaws in end to end Machine Learning debugging
MLOps.community
51 3rd wave of data scientists
3rd wave of data scientists
MLOps.community
52 MLOps meetup #7 Alex Spanos // TrueLayer 's MLOps Pipeline
MLOps meetup #7 Alex Spanos // TrueLayer 's MLOps Pipeline
MLOps.community
53 MLOps Meetup #8 Optimizing Your ML Workflow with Kubeflow 1.0
MLOps Meetup #8 Optimizing Your ML Workflow with Kubeflow 1.0
MLOps.community
54 Are Kubeflow and Airflow complementary?
Are Kubeflow and Airflow complementary?
MLOps.community
55 Why Kubeflow gained so much traction=open community
Why Kubeflow gained so much traction=open community
MLOps.community
56 Who decides the dirrection of Kubeflow
Who decides the dirrection of Kubeflow
MLOps.community
57 What do Kubeflow and Arrikto do and how do they work together?
What do Kubeflow and Arrikto do and how do they work together?
MLOps.community
58 Versioning your ML steps with Kubeflow
Versioning your ML steps with Kubeflow
MLOps.community
59 Machine Learning Lifecycles//Perception vs Reality
Machine Learning Lifecycles//Perception vs Reality
MLOps.community
60 Kubeflow vs SageMaker in Machine Learning
Kubeflow vs SageMaker in Machine Learning
MLOps.community

Chris Van Pelt shares advice on securing MLOps projects, including implementing least privilege, using good security tooling, and tracking vendors from the beginning. This is crucial for ensuring the security and compliance of MLOps projects.

Key Takeaways
  1. Implement least privilege for new engineers
  2. Use good security tooling such as scanning for CVEs
  3. Track vendors and their compliance
  4. Use dependabot for vulnerability management
  5. Get SOC 2 compliance for mature vendors
💡 Implementing least privilege and tracking vendors from the beginning can go a long way in ensuring the security and compliance of MLOps projects.

Related Reads

📰
A Phased Blueprint for Migrating From Google Workspace to Microsoft 365
Learn a step-by-step approach to migrate from Google Workspace to Microsoft 365 with minimal downtime and zero data loss, understanding it as an infrastructure engineering challenge
Hackernoon
📰
Feature Freshness: The Forgotten Problem of MLOps
Learn how outdated features can cause production models to fail and why feature freshness is crucial in MLOps, to improve model performance and reliability
Medium · LLM
📰
Day 19 of the 100 Days of MLOps Challenge
Learn to build a complete DVC ML pipeline with remote storage and experiments to streamline your machine learning workflow and improve collaboration
Medium · DevOps
📰
From Critical Infrastructure to AI Factories: Building an AI Operations Copilot on Nebius…
Learn how to build an AI operations copilot by leveraging experience in critical infrastructure and AI-assisted engineering, and why it matters for efficient AI deployment
Medium · LLM
Up next
Pole Pruner How A Rope Lever Shears High Branches
Innoforge Studio
Watch →