🚗 Driving Innovation: Machine Learning in Auto Claims Processing
Jul 15, 2024
auto_awesome
Michał Oleszak, an ML engineering manager at Solera, talks about using ML in auto claims processing, challenges in deploying ML pipelines, data quality for computer vision tasks, and exciting developments in self-supervised learning. He also discusses monorepo architecture benefits, model evaluation, and the importance of statistics in ML.
Machine learning aids in precise vehicle damage assessment for insurance claims processing.
Implementing a monorepo architecture optimizes code integration and model deployment for ML pipelines.
Deep dives
Machine Learning Solutions for Automotive Claims Processing
Solera's machine learning-based solutions for automotive claims processing involve a range of models that assist in various tasks, such as determining if a car is a total loss, estimating repair costs, and assessing repair offer fairness. By analyzing images of damaged vehicles, the models enable efficient decision-making for insurers and consumers, providing valuable insights into the necessary repairs and associated costs.
Challenges and Solutions in Model Deployment and Integration
The complexity of Solera's multi-stage pipeline for car damage assessment requires a thoughtful approach to integrating various models effectively. Combining deep learning models with rule-based systems, the deployment process is highlighted as integral from the inception of model development. Encouraging a codebase-centric deployment strategy and utilizing a robust CI/CD pipeline streamlines the integration of machine learning components.
Monorepo Architecture and Model Interface Standardization
Transitioning towards a monorepo architecture, Solera faces challenges in consolidating diverse code repositories and aligning different logical components within a unified system. Utilizing templates or contracts ensures a standardized approach for model interfaces, facilitating smoother code contributions and reducing development effort. Developing a consistent build system simplifies the merging process and enhances operational efficiency.
Data Quality Management and Validation Toolkit
Monitoring the quality of data inputs for machine learning tasks is crucial at Solera, with a focus on ensuring tabular data reliability and coherence. Leveraging tools such as Great Expectations enables automated data validation and facilitates clean data preparation for training and testing sets. Addressing objective and subjective image quality aspects allows for effective curation of datasets, tailoring them to specific model requirements for optimal performance.
In this episode, Dean speaks with Michał Oleszak, an ML engineering manager at Solera. Michał shares insights into how his team is using machine learning to transform the automotive claims process, from recognizing vehicle damages in images to estimating repair costs. The conversation covers the challenges of deploying ML pipelines in production, managing data quality for computer vision tasks, and balancing technical implementation with business needs. Michał also discusses his approach to model evaluation, the benefits of monorepo architecture, and his views on exciting developments in self-supervised learning for computer vision.
Join our Discord community: https://discord.gg/tEYvqxwhah
---
Timestamps:
00:00 Introduction
00:42 Production for Machine Learning at Solera
03:49 Transitioning from Images to Structured Data
04:58 Combining Deep Learning and Non-Deep Learning Models
05:15 Deployment Process for Machine Learning Models
08:01 Challenges and Solutions in Monorepo Adoption
12:57 Evaluating Model and Pipeline Versions
21:57 Tools for ML Projects: Monorepo, Pants, GitHub Actions
24:04 Data Management and Data Quality
30:14 Challenges in ML Efforts: Data Quality
30:37 Excitement about Self-Supervised Learning and JEPA Architectures
34:45 Controversial Opinion: Importance of Statistics for ML
36:40 Recommendations
Links
🌎Prisoners of Geography by Tim Marshall: https://www.amazon.com/Prisoners-Geography-Explain-Everything-Politics/dp/1501121472
➡️ Michał Oleszak on LinkedIn – https://www.linkedin.com/in/michal-oleszak/
➡️ Michał Oleszak on Twitter – https://x.com/MichalOleszak
🌐 Check Out Our Website! https://dagshub.com
Social Links:
➡️ LinkedIn: https://www.linkedin.com/company/dagshub
➡️ Twitter: https://twitter.com/TheRealDAGsHub
➡️ Dean Pleban: https://twitter.com/DeanPlbn
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode