NikoTak – Tamara Shostak's blog

As a data scientist, I witnessed a pivotal moment during the Wednesday morning keynote at Google Cloud Next ’18. Google announced BigQuery Machine Learning (BQML), and I immediately recognized its potential to transform how we approach security. Today, as I reflect on that moment through the lens of the transformer revolution sparked by the “Attention Is All You Need” paper, I’m more excited than ever about where we’re heading.

I was deeply involved in Reblaze’s participation in this announcement. When Google demonstrated Cloud Armor with a live DDoS attack on our protected site, I couldn’t help but smile – we were already exploring ways to combine BQML with attention-based models for threat detection, though that wasn’t mentioned in the demo. It was our little secret at the time, but now I can share my vision for what’s coming next.

Being part of the BQML closed pre-alpha was an incredible experience. Our use cases at Reblaze weren’t just pushing ML boundaries – they were showing me the path toward what I believe will be the future of security intelligence. As someone who lives and breathes data science, I could see the possibilities that others might have missed.

BQML democratized Machine Learning by making it as accessible as SQL. But I see something even more revolutionary on the horizon: combining BQML’s data processing capabilities with transformer architecture’s ability to understand complex patterns. I’m not just talking about analyzing billions of HTTP/S transactions – I envision an AI that understands the subtle contexts and intentions behind each request, just as I do when I’m analyzing security patterns.

The fundamental benefits of serverless Big Data capabilities are just the beginning:

• Unlimited data capture and access without infrastructure overhead
• Distributed storage with global accessibility
• SaaS advantages including cost-effectiveness and remote management

But what excites me most is how we can enhance these capabilities with transformer-based AI to:

• Learn and adapt from global attack patterns in real-time
• Understand the contextual relationships between seemingly unrelated security events
• Make predictive decisions based on subtle pattern recognition
• Scale intelligence across entire security infrastructures

Our single-tenant architecture at Reblaze, which gives each customer their own Virtual Private Cloud, is the perfect foundation for what I envision. While others struggle with multi-tenancy vulnerabilities, I see a future where each customer has their own dedicated security brain powered by transformer architecture.

When BQML first launched with support for Linear Regression and Binary Logistic Regression models, I already knew this was just the beginning. The transformer-based models I’m working on will:

• Process sequential security events with long-range dependencies
• Understand attack patterns across multiple time scales
• Adapt to emerging threats through continuous learning
• Share intelligence across instances while maintaining tenant isolation

This isn’t just a technological advancement – it’s a fundamental shift in how we approach security. Just as BQML eliminated the need for separate ML environments, I believe our integration of transformer architecture will create a unified intelligence layer spanning the entire security stack.

The future I’m building combines the best of both worlds:

• BQML’s ability to process massive amounts of security data
• Transformer architecture’s capability to understand complex patterns
• Real-time threat adaptation and response
• Automated intelligence that grows stronger with each interaction

For our customers, this means security that doesn’t just react to threats – it anticipates them. With each rolling update, I’m working to include more elements of this transformer-based intelligence, building toward the future I envision.

If you share my vision for the future of security, I’d love to show you what we’re working on at Reblaze. You can try it for free via Google Cloud Marketplace, where deployment takes just a few clicks. As a data scientist passionate about pushing the boundaries of what’s possible, I’m always excited to connect with others who see the potential in these technologies.