Improve Deep Learning Embeddings with Triplet Learning

A simple tweak to skyrocket results

4 min readJul 29, 2024

Trying to build embeddings in AI? Allow me to introduce you to a game-changer that will skyrocket your performance.

Embeddings are the unsung heroes of machine learning — compact numerical representations of data that make algorithms go “aha!” But getting embeddings right is tricky. Enter Triplet Learning, the secret weapon for embeddings that really understand what’s similar and what’s not.

Why Triplet Learning? It’s All About the Triad

The problem with some embeddings is that they focus on absolute distance, not what matters: relative similarity. Triplet Learning fixes this by using a trio of data points-

Anchor: Your starting point, like a picture of a cat.

Positive: Something that should be close to your anchor, like another cat pic.

Negative: Something that should be far from your anchor, like a picture of a dog.

Triplet Learning’s loss function two simple goals:

Shove together: Anchor and positive embeddings

Push apart: Anchor and negative embeddings

This forces the model to learn that even if two cat pictures look different (different breeds, poses), they’re closer to each other than either is to the dog. This is huge for stuff like face recognition or recommendation systems. An example is shown in the image below. We train the CNN to embed the two pictures of notable Basketball Fan Barak Obama together, and the negative sample further away. In doing so it can recognize Obama in different angles, lighting, expressions etc.

Why It Works: The Power of Three

Relativity Rules: We don’t care about how far apart things are in some abstract sense, just that similar things are closer than dissimilar things. Triplet Loss gets this.

Flexible Friend: Whether it’s images, text, or neural networks, Triplet Learning doesn’t discriminate. It’s your adaptable pal for better embeddings.

See It To Believe It: Ever wanted to visualize why two things are considered similar? With Triplet Learning’s embeddings, you can often see the clusters forming.

Triplet Learning In Action: Use Cases

Facial Recognition Done Right: Remember when the iPhone couldn’t tell Chinese people apart? Triplet Learning can help avoid those awkward mix-ups.

Recommendations You’ll Actually Like: Triplet-powered embeddings get what you’re into and suggest things you’ll genuinely enjoy.

Find Your Match (In Data): Image and text retrieval become way more accurate when your embeddings truly understand similarity.

Bottom Line

If your embeddings are struggling to understand the concept of “alike,” Triplet Learning is your solution. By focusing on the relative relationships between three points, you get embeddings that are actually useful for real-world tasks. It’s a simple idea with a big impact.

That’s why I’m a huge fan of Triplet Learning. In-fact it is one of the pillars of our cutting-edge Deepfake system here-

Deepfake Detection: Building the Future with Advanced AI Techniques [Deepfakes]

How to Detect Deepfakes with AI Part 2: A Complete Pipeline

artificialintelligencemadesimple.substack.com

If you liked this article and wish to share it, please refer to the following guidelines.

That is it for this piece. I appreciate your time. As always, if you’re interested in working with me or checking out my other work, my links will be at the end of this email/post. And if you found value in this write-up, I would appreciate you sharing it with more people. It is word-of-mouth referrals like yours that help me grow.

I put a lot of effort into creating work that is informative, useful, and independent from undue influence. If you’d like to support my writing, please consider becoming a paid subscriber to this newsletter. Doing so helps me put more effort into writing/research, reach more people, and supports my crippling chocolate milk addiction. Help me democratize the most important ideas in AI Research and Engineering to over 100K readers weekly.

Help me buy chocolate milk

PS- We follow a “pay what you can” model, which allows you to support within your means. Check out this post for more details and to find a plan that works for you.

I regularly share mini-updates on what I read on the Microblogging sites X(https://twitter.com/Machine01776819), Threads(https://www.threads.net/@iseethings404), and TikTok(https://www.tiktok.com/@devansh_ai_made_simple)- so follow me there if you’re interested in keeping up with my learnings.