Input → [Update Gate] → [Hidden State] → Output
        ⬆️           ⬇️
       [Reset Gate] → (Controls memory update)

Imagine you are reading a story.

You don’t need to remember every word, just the important details.

💡 GRUs are like LSTMs but simpler and faster!

They remember important things and forget unnecessary ones, helping computers understand speech, text, and time-series data better.

1. Why Do We Need GRUs?

Problem with RNNs: They Forget! ❌

• RNNs lose important information in long sequences (e.g., remembering a story).
• LSTMs fix this but are slow and complex.

🔹 GRUs = A Simpler, Faster LSTM!

• They work like LSTMs but have fewer gates, making them easier to train.
• LSTMs have 3 gates, but GRUs have only 2 gates!

2. How Do GRUs Work?

🔹 GRUs use 2 gates:

💡 GRUs decide what to forget and what to keep dynamically.

3. How GRUs Work Mathematically (Simplified)

At every step tt:

1️⃣ Reset Gate (Forget Old Info)

$$ r_t = \sigma(W_r \cdot [h_{t-1}, x_t]) $$