Proactive Steps for Securing MLOps Projects // Chris Van Pelt // MLOps Podcast #192 clip
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
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Our 1st MLOps Meetup // Luke Marsden // MLOps Meetup #1
MLOps.community
Remote Collaboration as a Data Scientist
MLOps.community
MLOps Manifesto with Luke Marsden from Dotscience
MLOps.community
MLOps lifecycle description
MLOps.community
What Does Best in Class AI/ML Governance Look Like in Fin Services? // Charles Radclyffe // MLOps #2
MLOps.community
Life purpose and too many spreadsheets
MLOps.community
Explainability, Black boxes and EU white paper on reproducibility
MLOps.community
Hierarchy of Machine Learning Needs // Phil Winder // MLOps Meetup #3
MLOps.community
Automatically Retrain Machine Learning Models? Are best practices worth it?
MLOps.community
Building an MLOps Team? Key ideas to keep in mind
MLOps.community
Hierarchy of MLOps Needs
MLOps.community
Bare necessities for getting an ML model into production
MLOps.community
MLOps and Monitoring
MLOps.community
How Phil Winder got into Data Science and Software Engineering
MLOps.community
Provenance and Reproducibility in Machine Learning; what is it and why you need it?
MLOps.community
Friction Between Data Scientists and Software Engineers
MLOps.community
MLOps Problems in different size companies
MLOps.community
ML tooling in large companies
MLOps.community
ML Platforms - The build vs buy question
MLOps.community
ML Services Gateway at SurveyMonkey
MLOps.community
Message buses, Async and sync architecture
MLOps.community
MLOps #4: Shubhi Jain - Building an ML Platform @SurveyMonkey
MLOps.community
Hybrid Data Science Teams @SurveyMonkey
MLOps.community
How do you handle ML version control at SurveyMonkey
MLOps.community
Doing ML with Personal Information
MLOps.community
Evolution of the ML feature store @SurveyMonkey
MLOps.community
Developing a Machine Learning Feature Store
MLOps.community
Auto retrain ML models is not the question
MLOps.community
3 key parts to Machine Learning monitoring
MLOps.community
MLOps Meetup #6: Mid-Scale Production Feature Engineering with Dr. Venkata Pingali
MLOps.community
MLOps meetup #5 High Stakes ML: Active Failures, Latent Factors with Flavio Clesio
MLOps.community
MLOps: Airflow Pros and Cons
MLOps.community
Specific challenges in Machine Learning
MLOps.community
Current State Of Machine Learning
MLOps.community
Humans in the Loop are a defining factor in Machine Learning
MLOps.community
Learning from real life Machine Learning failures
MLOps.community
Survivorship Bias in machine learning tutorials
MLOps.community
Swiss Cheese model in Machine Learning
MLOps.community
Resume driven development in Machine learning & software engineering
MLOps.community
Who has the highest standards in ML?
MLOps.community
Venkata Pingali of Scribble Data Thoughts on the Current State of Machine Learning
MLOps.community
Dependable data and being able to Trust in your Data with Venkata Pengali of Scribble Data
MLOps.community
Speed, Trust, Evolution and Scale in MLOps
MLOps.community
More difficult transition for data scientists to become ML engineers
MLOps.community
How many models in prod til I need a dedicated ML platform?
MLOps.community
Deeper thinking from data scientists around platform blackholes
MLOps.community
Checkpointing, metadata, and confidence in your data
MLOps.community
Adjacent usecases and multistep feature engineering
MLOps.community
Standardization of Machine Learning tools like in Software Engineering with Venkata Pingali
MLOps.community
Reproducability flaws in end to end Machine Learning debugging
MLOps.community
3rd wave of data scientists
MLOps.community
MLOps meetup #7 Alex Spanos // TrueLayer 's MLOps Pipeline
MLOps.community
MLOps Meetup #8 Optimizing Your ML Workflow with Kubeflow 1.0
MLOps.community
Are Kubeflow and Airflow complementary?
MLOps.community
Why Kubeflow gained so much traction=open community
MLOps.community
Who decides the dirrection of Kubeflow
MLOps.community
What do Kubeflow and Arrikto do and how do they work together?
MLOps.community
Versioning your ML steps with Kubeflow
MLOps.community
Machine Learning Lifecycles//Perception vs Reality
MLOps.community
Kubeflow vs SageMaker in Machine Learning
MLOps.community
More on: AI Security
View skill →Related Reads
📰
📰
📰
📰
A Phased Blueprint for Migrating From Google Workspace to Microsoft 365
Hackernoon
Feature Freshness: The Forgotten Problem of MLOps
Medium · LLM
Day 19 of the 100 Days of MLOps Challenge
Medium · DevOps
From Critical Infrastructure to AI Factories: Building an AI Operations Copilot on Nebius…
Medium · LLM
🎓
Tutor Explanation
DeepCamp AI