http://shivasurya.me/categories/neural-networks/Recent content in Neural-Networks on ShivasuryaHugoen-usFri, 08 Aug 2025 00:00:00 +0000http://shivasurya.me/2025/08/08/neural-network/Fri, 08 Aug 2025 00:00:00 +0000http://shivasurya.me/2025/08/08/neural-network/<p>Back in 2017, I used to tease my friend about his machine learning work (training models, dataset operations, ML deployments) - &ldquo;Come on, admit it, aren&rsquo;t you just writing complex if-elif-else statements and calling yourself an ML engineer?&rdquo;. While Google was <a href="https://ai.google.dev/edge/litert/android">bringing ML models</a> to mobile devices using tensorflow, I remained indifferent as I couldn&rsquo;t grasp the underlying mathematics or internal workings (which, honestly, continues to this day).</p> <p><img src="http://shivasurya.me/assets/media/if-else-ml.png" alt="IF-ELSE-Engineer">{:height=&ldquo;400px&rdquo;}</p> <p>My perspective shifted after diving deep into projects like <a href="https://shivasurya.me/llm/ai/2025/04/10/lessons-from-building-sherlock-automating-security-code-reviews-with-sourcegraph.html">Sherlock</a>, <a href="https://shivasurya.me/llm/ai/2025/03/19/llm-powered-security-reviews.html">LLM-Powered Security Reviews</a>, and <a href="https://codepathfinder.dev/blog/introducing-secureflow-extension-to-vibe-code-securely/">SecureFlow AI</a>. After studying numerous research papers about using language models to detect code vulnerabilities, I felt compelled to return to the basics and understand the fundamental workings of neural networks, particularly how they store information in their hidden layers.</p>