Self-healing neural networks via modular patch layers for diverse structural and adversarial damages
B. Santhosh Reddy, S. Deepa Nivethika, B. Vara Anjan, P. Mahidhar Reddy, M. SenthilPandian, R. Roobankumar · 2026
Read Self-healing neural networks via modular patch layers for diverse structural and adversarial damages by B. Santhosh Reddy, S. Deepa Nivethika, B. Vara Anjan, P. Mahidhar Reddy, M. SenthilPandian, R. Roobankumar (2026) 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/PMC13572459.txt.
Description
B. Santhosh Reddy, with five colleagues, presents a 2026 design for neural networks that can repair themselves after damage. Networks in safety-linked systems can suffer broken inner parts or attacks that feed them crafted inputs. Full retraining is slow and costly, so the design aims for small, local fixes instead.
The model scans each layer for odd activity, finds the faulty layer, and retrains only a light patch for that spot. Tests ran on four image data sets, from hand digits to street-style photos. After parts were zeroed out, accuracy bounced back near the intact level, at 97.33 and 97.22 percent against a 97.76 percent base in one case. The repair adds little compute or memory load. The authors pitch it for systems that must keep running without a full rebuild.
Catalog record
| CREATOR | B. Santhosh Reddy, S. Deepa Nivethika, B. Vara Anjan, P. Mahidhar Reddy, M. SenthilPandian, R. Roobankumar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 116894 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |