# long short-term memory

> artificial recurrent neural network architecture used in deep learning

**Wikidata**: [Q6673524](https://www.wikidata.org/wiki/Q6673524)  
**Wikipedia**: [English](https://en.wikipedia.org/wiki/Long_short-term_memory)  
**Source**: https://4ort.xyz/entity/long-short-term-memory

## Summary
Long short-term memory (LSTM) is an artificial recurrent neural network architecture used in deep learning. It is a type of recurrent neural network (RNN) introduced in 1995 by Sepp Hochreiter and Jürgen Schmidhuber and is commonly referred to by the short name "LSTM."

## Key Facts
- LSTM is a subclass of recurrent neural network, a class of artificial neural network where connections between units form a directed graph along a temporal sequence.  
- Inventors: Sepp Hochreiter and Jürgen Schmidhuber.  
- Time of discovery / earliest written record: 1995.  
- Short name / alias: LSTM (also long short-term memory and multiple multilingual aliases such as 长短期记忆神经网络).  
- Defining formula (gate and state update equations) is commonly presented as:
  f_t = σ_g(W_f x_t + U_f h_{t-1} + b_f)  
  i_t = σ_g(W_i x_t + U_i h_{t-1} + b_i)  
  o_t = σ_g(W_o x_t + U_o h_{t-1} + b_o)  
  \tilde{c}_t = σ_c(W_c x_t + U_c h_{t-1} + b_c)  
  c_t = f_t ∘ c_{t-1} + i_t ∘ \tilde{c}_t  
  h_t = o_t ∘ σ_h(c_t)  
- Representative schematics available at: The_LSTM_Cell.svg, Peephole_Long_Short-Term_Memory.svg, and Long_Short-Term_Memory_cell,_1997_version.png (Wikimedia Commons file paths).  
- Identifier data: golden_id Long_short-term_memory-6WPR4M; freebase_id /m/02qmzyq; microsoft_academic_id_(discontinued) 133488467.  
- Medical Subject Headings (MeSH) tree codes: G17.485.937.750 and L01.224.050.375.605.937.750 (qualifier: recurrent neural network).  
- Wikipedia title: "Long short-term memory"; sitelink_count: 26 across languages (examples: ar, bn, ca, cs, de, el, en, es, fa, gl).  
- Maintained by: WikiProject Mathematics (Wikipedia maintenance association).

## FAQs
### Q: What does LSTM stand for?
A: LSTM stands for "long short-term memory." It is commonly used as the short name for the architecture.

### Q: Who invented LSTM and when?
A: LSTM was invented by Sepp Hochreiter and Jürgen Schmidhuber, with the time of discovery or earliest written record dated 1995.

### Q: How is an LSTM cell defined mathematically?
A: An LSTM cell is defined by gate and state update equations that compute forget gate f_t, input gate i_t, output gate o_t, candidate cell state \tilde{c}_t, updated cell state c_t, and output hidden state h_t as shown in the defining formula (see Key Facts).

## Why It Matters
LSTM is significant because it is a distinct architecture within the class of recurrent neural networks and is explicitly documented and identified in the deep learning literature. Introduced in 1995 by Sepp Hochreiter and Jürgen Schmidhuber, LSTM formalizes a set of gate and state-update equations that define how information is carried and transformed across time steps in a recurrent architecture. The architecture is represented by canonical schematics and has dedicated identifiers in academic and knowledge databases (freebase, MeSH, Microsoft Academic). Its prominence is reflected by multilingual coverage and numerous linked resources (Wikipedia sitelinks and Wikimedia Commons schematics). As a named, well-defined RNN variant, LSTM functions as a reference architecture in discussions and implementations of recurrent models in deep learning.

## Notable For
- Being a formally defined RNN architecture introduced in 1995 by Sepp Hochreiter and Jürgen Schmidhuber.  
- Having an explicit, widely cited set of gate and state-update equations (the defining formula).  
- Possessing multiple canonical schematic representations available on Wikimedia Commons (including a 1997 cell diagram and variants).  
- Being indexed across multiple knowledge systems and identifiers (golden_id, freebase_id, MeSH codes, Microsoft Academic id).

## Body

### Overview
- Long short-term memory (LSTM) is an artificial recurrent neural network architecture used in deep learning.  
- It is categorized as a subclass of recurrent neural network — a class where connections form a directed graph along a temporal sequence.

### History and provenance
- Discoverers / inventors: Sepp Hochreiter and Jürgen Schmidhuber.  
- Time of discovery / earliest written record: 1995 (as recorded in the structured properties).  
- The entity is maintained and documented on Wikipedia under the title "Long short-term memory" and is associated with WikiProject Mathematics for editorial maintenance.

### Architecture and defining equations
- The canonical LSTM cell is described by the following equations:
  f_t = σ_g(W_f x_t + U_f h_{t-1} + b_f)  
  i_t = σ_g(W_i x_t + U_i h_{t-1} + b_i)  
  o_t = σ_g(W_o x_t + U_o h_{t-1} + b_o)  
  \tilde{c}_t = σ_c(W_c x_t + U_c h_{t-1} + b_c)  
  c_t = f_t ∘ c_{t-1} + i_t ∘ \tilde{c}_t  
  h_t = o_t ∘ σ_h(c_t)
- These equations define gates (f_t, i_t, o_t), a candidate cell state (\tilde{c}_t), the cell state (c_t), and the hidden/output state (h_t).

### Schematics and visual resources
- Representative images and schematics are available (file paths on Wikimedia Commons):  
  - The_LSTM_Cell.svg (English)  
  - Peephole_Long_Short-Term_Memory.svg (English)  
  - Long_Short-Term_Memory_cell,_1997_version.png

### Identifiers and classification
- Short name / alias: LSTM; other multilingual aliases included in the structured aliases list.  
- golden_id: Long_short-term_memory-6WPR4M.  
- freebase_id: /m/02qmzyq.  
- microsoft_academic_id_(discontinued): 133488467.  
- MeSH tree codes: G17.485.937.750 and L01.224.050.375.605.937.750 (linked to recurrent neural network).  
- mesh_descriptor_id: D000098417 ("Long Short Term Memory").

### Related persons
- Sepp Hochreiter — German computer scientist; occupations listed as bioinformatician, computer scientist, artificial intelligence researcher; birth date: 1967-02-14.  
- Jürgen Schmidhuber — German computer scientist and artificial intelligence researcher; occupations include computer scientist and university teacher; birth date: 1963-01-17; citizenship: Germany.

### Documentation and community tags
- Wikipedia languages: ar, bn, ca, cs, de, el, en, es, fa, gl, etc. (sitelink_count: 26).  
- Stack Exchange / community tags: Stack Overflow and AI Stack Exchange topic tags for LSTM / long-short-term-memory.  
- GitHub topics: long-short-term-memory, lstm.

## References

1. Long Short Term Memory
2. [OpenAlex](https://docs.openalex.org/download-snapshot/snapshot-data-format)