deep-learning.html

<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Era Details: Deep Learning</title>
    <link rel="stylesheet" href="style.css">
    <style>
        .details-body { max-width: 800px; margin: 0 auto; padding: 40px 20px; text-align: left; }
        .hero-section { border-bottom: 2px solid var(--primary); padding-bottom: 20px; margin-bottom: 40px; }
        .quote { font-style: italic; font-size: 1.5rem; color: #ef4444; margin: 10px 0; }
        .back-btn { display: inline-block; margin-bottom: 20px; color: var(--primary); text-decoration: none; font-weight: bold; }
        
        .sub-timeline { border-left: 2px solid #334155; padding-left: 20px; margin-left: 10px; }
        .milestone { margin-bottom: 30px; position: relative; }
        .milestone::before { content: ""; position: absolute; left: -26px; top: 5px; width: 10px; height: 10px; background: #ef4444; border-radius: 50%; }
        
        .methodology-box { background: #1e293b; border-left: 4px solid #ef4444; padding: 20px; margin: 20px 0; }
        .architecture-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(250px, 1fr)); gap: 20px; }
        .arch-card { background: #161e2e; border: 1px solid #334155; padding: 15px; border-radius: 8px; }
    </style>
</head>
<body>
    <div class="details-body">
        <a href="index.html" class="back-btn">← Back to Timeline</a>
        
        <div class="hero-section">
            <span class="year" style="color: #ef4444;">2011 – 2020</span>
            <h1>The Deep Learning Era</h1>
            <p class="quote">"The network is the feature: Learning layers of abstraction."</p>
        </div>

        <section>
            <h2>Chronology of the Neural Revolution</h2>
            <div class="sub-timeline">
                <div class="milestone">
                    <h4>2012: The AlexNet Breakthrough</h4>
                    <p>Alex Krizhevsky and Geoffrey Hinton win the ImageNet competition by a landslide using a Deep CNN. This proved that GPUs and Deep Nets were the future.</p>
                </div>
                <div class="milestone">
                    <h4>2014: GANs (Generative Adversarial Networks)</h4>
                    <p>Ian Goodfellow introduces GANs, where two networks compete. One creates images, the other critiques them. This was the birth of AI-generated art.</p>
                </div>
                <div class="milestone">
                    <h4>2016: AlphaGo Defeats Lee Sedol</h4>
                    <p>Google DeepMind's AlphaGo defeats the world champion in Go, a game previously thought impossible for AI due to its infinite complexity.</p>
                </div>
                <div class="milestone">
                    <h4>2017: The "Attention is All You Need" Paper</h4>
                    <p>Google researchers introduce the **Transformer** architecture. This eventually replaced RNNs and paved the way for modern LLMs.</p>
                </div>
            </div>
        </section>

        <section style="margin-top: 40px;">
            <h2>Core Architectures (The "Species" of AI)</h2>
            <div class="architecture-grid">
                <div class="arch-card">
                    <h4>CNNs (Convolutional)</h4>
                    <p>Designed for spatial data like images. They use "filters" to detect edges, then shapes, then objects.</p>
                </div>
                <div class="arch-card">
                    <h4>RNNs & LSTMs</h4>
                    <p>Designed for sequential data like speech and text. They have "memory" of what happened in the previous step.</p>
                </div>
                <div class="arch-card">
                    <h4>Autoencoders</h4>
                    <p>Used for compression and noise removal. They learn to reconstruct their input from a condensed "bottleneck" layer.</p>
                </div>
                <div class="arch-card">
                    <h4>Deep Reinforcement Learning</h4>
                    <p>Combining neural nets with trial-and-error rewards. Used for robotics and mastering video games.</p>
                </div>
            </div>
        </section>
        <section style="margin-top: 40px; border: 1px solid #334155; padding: 25px; border-radius: 12px; background: linear-gradient(180deg, #161e2e, #0f172a);">
            <h2 style="color: #ef4444;">The Hardware Catalyst: From Silicon to Neural Engines</h2>
            <p>AI reached a "level up" not just because of better code, but because we changed the physical way computers think.</p>
            
            <div class="tech-grid">
                <div class="tech-item">
                    <h4>1. The GPU Revolution (Nvidia Shift)</h4>
                    <p><strong>The Move:</strong> Moving from CPU to GPU. While a CPU handles a few complex tasks in a row (Serial), a GPU handles thousands of simple math tasks at once (Parallel).</p>
                    <p><em>Why it matters:</em> Neural networks are just massive matrices of multiplication. GPUs can do billions of these per second.</p>
                </div>
                <div class="tech-item">
                    <h4>2. CUDA & Software Abstraction</h4>
                    <p><strong>The Move:</strong> Nvidia’s CUDA allowed researchers to write C++ code directly for the GPU. This turned a "video card" into a general-purpose AI brain.</p>
                </div>
                <div class="tech-item">
                    <h4>3. TPU (Tensor Processing Units)</h4>
                    <p><strong>The Move:</strong> Google developed ASICs (Application-Specific Integrated Circuits) designed <em>specifically</em> for the matrix math used in AI, stripping away everything a computer doesn't need for neural nets.</p>
                </div>
                <div class="tech-item">
                    <h4>4. HBM (High Bandwidth Memory)</h4>
                    <p><strong>The Move:</strong> The bottleneck wasn't just calculation speed; it was moving data to the chip. HBM allowed "stacks" of memory to sit right next to the processor, providing the speed needed for LLMs.</p>
                </div>
            </div>
        </section>
        <section style="margin-top: 40px;">
            <h2>Methodologies & Optimization Techniques</h2>
            <div class="methodology-box">
                <h4>1. Backpropagation & Stochastic Gradient Descent (SGD)</h4>
                <p>The mathematical engine. The model calculates its "error" and sends it backward through the network to update millions of weights using <strong>Calculus (Derivatives)</strong>.</p>
            </div>
            <div class="methodology-box">
                <h4>2. Activation Functions (ReLU, Softmax)</h4>
                <p>Non-linear functions that decide if a neuron should "fire." <strong>ReLU</strong> solved the "vanishing gradient" problem, allowing nets to be 100+ layers deep.</p>
            </div>
            <div class="methodology-box">
                <h4>3. Transfer Learning</h4>
                <p>The methodology of taking a model trained on one giant task (like recognizing cats) and "fine-tuning" it for a specific task (like detecting cancer in X-rays).</p>
            </div>
            <div class="methodology-box">
                <h4>4. Dropout & Batch Normalization</h4>
                <p>Methods used during training to keep the network stable and prevent it from becoming overly reliant on specific "pathways," ensuring better generalization.</p>
            </div>
        </section>

        <section style="margin-top: 40px;">
            <h2>The Approach: End-to-End Learning</h2>
            <p>The fundamental approach shifted to <strong>End-to-End</strong>. You no longer tell the AI to look for "eyes" and "ears" to find a face. You give it 10 million faces, and it discovers that "eyes" are a statistically significant pattern on its own. <strong>The machine builds its own internal dictionary.</strong></p>
        </section>
    </div>
</body>
</html>