Understanding Vertical and Horizontal Full Stack Machine Learning Engineering

There are two aspects to Full Stack Machine Learning (ML) Engineering – vertical and horizontal. These represent different dimensions of the field and both are vital to achieving success in ML projects.

The Vertical Side of Full Stack ML

Vertical Full Stack ML encompasses the end-to-end execution of machine learning from data collection to deployment. Landing AI established a complete methodology, the “Machine Learning Lifecycle,” that details every stage of the ML process. Here’s a brief overview:

  1. Data Collection: Creating a suitable data collection pipeline is crucial. For instance, in computer vision, setting up the correct camera system and lighting can affect the quality of input photos or videos.
  2. Data Labeling: This step involves defining a labeling schema and the actual data labeling. A thorough labeling book, prepared with the help of labelers, serves as the cornerstone for its models.
  3. Model Iteration: Multiple model training jobs are launched, and a systemic error analysis is performed. This iterative process continues until the desired performance metrics are achieved.
  4. Continuous Deployment: Models are deployed and kept running in production, using a CI/CD pipeline and Over-the-Air solutions for deploying new inference code and models.
  5. Real-time Monitoring: Once deployed, continuous monitoring is essential to promptly detect any potential issues like model performance drift, latency, or overheating.

However, specialization often leads to a siloed approach. The key to success is a holistic view of the entire pipeline, enabling the identification of potential bottlenecks or issues in any component of the process.

The Horizontal Side of Full Stack ML

Horizontal Full Stack ML, though less discussed, is equally important. It involves being proficient across various types of ML, from supervised learning and meta-learning to generative models and graph models. It also includes having knowledge across multiple domains, from computer vision and Natural Language Processing (NLP) to recommender systems and reinforcement learning.

The essence of horizontal full stack ML is to maintain an open mind and extend your knowledge as widely as possible, moving beyond your comfort zone. A diverse knowledge base in ML can foster new techniques and fresh ideas.

Often, companies tend to hire experts in specific fields. This can lead to professionals focusing on a single area, ignoring other domains. However, such a trend can hamper the growth of ML, as breakthroughs often occur at the intersection of disciplines. Real-world data flow in a variety of formats, and limiting ourselves to one type of data format might lead to missed opportunities.

Landing AI promotes the concept of “life-long learning,” encouraging our team to explore all fields of ML. This open-mindedness is essential for solving real-world problems, for which no pre-existing playbook or boundaries exist. We should not limit ourselves to familiar methods, but rather, become full-stack horizontally to find the most optimal solutions.

In conclusion, a holistic and diverse approach to Full Stack ML, both vertically and horizontally, can optimize problem-solving and potentially lead to new breakthroughs in the field of Machine Learning.

Source:
Two Types of Full Stack Machine Learning Engineering


Get in touch

Whether you’re looking for expert guidance on an AI initiative or want to share your AI knowledge with others, our network is the place for you. Let’s work together to build a brighter future powered by AI.