EchoLSTM: A Self-Reflective Recurrent Network for Stabilizing Long-Range Memory
Prasanth K. K, Shubham Sharma · 2025
Read EchoLSTM: A Self-Reflective Recurrent Network for Stabilizing Long-Range Memory by Prasanth K. K, Shubham Sharma (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01950.txt.
Description
Prasanth K K and Shubham Sharma rebuild the memory of a recurrent network around self-reflection. Long short-term memory networks still lose long-range dependencies, especially when sequences contain distracting or misleading material. The authors' principle, output-conditioned gating, lets the model modulate its own memory gates using its past inferences, a feedback loop meant to stabilize what it retains.
Their EchoLSTM adds an attention mechanism on top of that principle. On a custom distractor signal task it reached about 69 percent accuracy, beating a standard LSTM baseline by 33 percentage points. On the ListOps benchmark it scored about 70 percent, close to a modern Transformer's roughly 72 percent, while using over five times fewer parameters. A trigger sensitivity test adds qualitative evidence that the self-reflective loop makes the memory fundamentally more robust.
Catalog record
| CREATOR | Prasanth K. K, Shubham Sharma |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 22 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-01950