Long Short-Term Memory (LSTM)
Definition
LSTM is a recurrent network architecture with gates designed to better preserve long-term dependencies than simple RNNs.
In today’s common variants, the memory cell as well as input, forget, and output gates control the information flow. LSTMs have been widely used especially for speech and time series.
Simply explained
Like a recurrent memory with controllers for retaining, forgetting, and outputting.
Why it matters in practice
Relevant for classical sequence and time series models.
Example from business
A forecasting model processes historical measurements with an LSTM and is compared against GRU and Transformer.
Limits and pitfalls
- Choosing LSTM uncritically for every time series; long sequences without benchmarking; ignoring scaling and data volume.
Origin & history
Hochreiter and Schmidhuber published LSTM in 1997.
Use in mid-sized companies
- Possible application in SMEs: Use as a sequence baseline when a recurrent architecture fits the task.
Frequently asked questions
Is LSTM a Transformer?
No. Is it always better than GRU? No.
Sources
- Long Short-Term Memory · Hochreiter & Schmidhuber (1997), Long Short-Term Memory
- Ergänzender Primärbeleg · Ergänzender Primärbeleg
Sources checked: 10./12.10.2026.
Editorially reviewed on: