Why I’m Beginning This Project
Right now, in October 2024, I’m launching a research project built around a simple but provocative question: Can network behavior be modeled the same way we model language?
Not as static events. Not as signatures. But as sequences with structure, grammar, and predictability.
Network traffic has patterns. It has transitions. It has intent. I want to treat it like a language and see what emerges.
My R&D Objective
This is not a production initiative. This is pure exploration. I’m building a pipeline that:
- Compresses raw traffic into symbolic tokens
- Learns transitions between those tokens
- Predicts the next state in a behavioral sequence
A forecasting engine, not a detection engine. My goal is to understand whether network behavior can be predicted when represented correctly.
The Energy Behind the Work
I’m deep in books, papers, and model architectures. Sequence modeling, NLP tokenization, Bayesian transitions, LSTMs, GRUs, CNNs, Transformers, I’m sweeping through everything I can find.
Each architecture offers a different lens:
- Bayesian models for clean probabilistic transitions
- MLPs as a grounding baseline
- RNNs for temporal memory
- LSTMs and GRUs for stability
- 1D CNNs for local temporal structure
- Transformers for long‑range dependencies
It feels like standing at the edge of a new field with a full toolbox and no constraints 🙂
Why This Matters Right Now
Traditional network analysis is reactive. Rules and signatures only tell you what already happened. Sequence modeling is proactive. It asks: What comes next?
That shift is the entire reason I’m doing this.
“Studying behavior is always more powerful than studying events.”
My Working Hypothesis
If I encode network behavior properly, compressed, tokenized, structured, then models should learn its patterns the same way they learn language. If they can learn language, they can learn traffic.
That’s the hypothesis I’m testing. And today I’m excited to begin 🙂