Introduction In the previous issue, we used the example of smartwatches to help you quickly understand how LSTM manages health data like a “memory master.” But did you know? Behind the power of LSTM lies a set of intricate mathematical formulas! Today, we will break down these formulas in the simplest language, allowing you to understand how each line of “code” drives the memory magic of LSTM!
The Four Key Formulas of LSTM: Dissecting the “Memory Algorithm”
The core of LSTM consists of four formulas that, like the algorithms of smartwatches, precisely control the entire process of “forgetting, remembering, updating, and outputting.” We will dissect these formulas line by line using the scenario of smartwatch monitoring heart rate!
1. Forget Gate: Deciding What to Forget
2. Input Gate: Deciding What to Remember
3. Memory Cell: The “Health Record” of Long-Term Memory
4. Output Gate: Deciding What to Output
The “Memory Philosophy” Behind the Formulas
The design of LSTM’s formulas reflects three core ideas:
- Selective Memory: Through gating mechanisms (forget gate, input gate), only important information is retained.
- Long-Term Dependency: The memory cell Ct acts like a “time capsule,” preserving key patterns over the long term.
- Dynamic Output: The output gate ot adjusts the output content flexibly based on current needs.
LSTM vs Human Memory: Striking Similarities
| LSTM Component | Human Memory Behavior |
|---|---|
| Forget Gate | Actively forgetting irrelevant details (e.g., “What did I eat yesterday?”) |
| Input Gate | Focusing on recording key events (e.g., “First time riding a bike”) |
| Memory Cell | Long-term preservation of important memories (e.g., “Graduation ceremony”) |
| Output Gate | Extracting memories based on context (e.g., “Telling an inspiring story during an interview”) |
Conclusion: The Formulas are the “Soul” of LSTM
Although the formulas of LSTM may seem complex, the underlying logic is simple: like humans, it selectively remembers, forgets, and outputs. Next time you see a “health alert” from a smartwatch, consider that the LSTM algorithm behind it may be executing a more sophisticated “memory management” process than humans do!
I hope this article has given you a more intuitive understanding of LSTM’s formulas! The formulas are the “soul” of LSTM, and you have already grasped its “thinking code”!