Activation functions in non-negative neural networks
Optical neural networks (ONNs) have the potential to overcome scaling limitations of transistor-based systems due to their inherent low latency and large available bandwidth. However, encoding the information directly in the physical properties of light fields also imposes new computational constrai...
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Dokumenttyp: | Article (Journal) |
| Sprache: | Englisch |
| Veröffentlicht: |
28 October 2025
|
| In: |
IEEE access
Year: 2025, Jahrgang: 13, Pages: 182474-182480 |
| ISSN: | 2169-3536 |
| DOI: | 10.1109/ACCESS.2025.3622408 |
| Online-Zugang: | Verlag, kostenfrei, Volltext: https://doi.org/10.1109/ACCESS.2025.3622408 Verlag, kostenfrei, Volltext: https://ieeexplore.ieee.org/document/11205509 
|
| Verfasserangaben: | Marlon Becker, Dominik Drees, Frank Brückerhoff-Plückelmann, Carsten Schuck, Wolfram Pernice, and Benjamin Risse |
MARC
| LEADER | 00000naa a2200000 c 4500 | ||
|---|---|---|---|
| 001 | 1962407578 | ||
| 003 | DE-627 | ||
| 005 | 20260224122506.0 | ||
| 007 | cr uuu---uuuuu | ||
| 008 | 260224s2025 xx |||||o 00| ||eng c | ||
| 024 | 7 | |a 10.1109/ACCESS.2025.3622408 |2 doi | |
| 035 | |a (DE-627)1962407578 | ||
| 035 | |a (DE-599)KXP1962407578 | ||
| 040 | |a DE-627 |b ger |c DE-627 |e rda | ||
| 041 | |a eng | ||
| 084 | |a 29 |2 sdnb | ||
| 100 | 1 | |a Becker, Marlon |d 1994- |e VerfasserIn |0 (DE-588)1364852837 |0 (DE-627)192494152X |4 aut | |
| 245 | 1 | 0 | |a Activation functions in non-negative neural networks |c Marlon Becker, Dominik Drees, Frank Brückerhoff-Plückelmann, Carsten Schuck, Wolfram Pernice, and Benjamin Risse |
| 264 | 1 | |c 28 October 2025 | |
| 300 | |b Illustrationen | ||
| 300 | |a 7 | ||
| 336 | |a Text |b txt |2 rdacontent | ||
| 337 | |a Computermedien |b c |2 rdamedia | ||
| 338 | |a Online-Ressource |b cr |2 rdacarrier | ||
| 500 | |a Gesehen am 24.02.2026 | ||
| 520 | |a Optical neural networks (ONNs) have the potential to overcome scaling limitations of transistor-based systems due to their inherent low latency and large available bandwidth. However, encoding the information directly in the physical properties of light fields also imposes new computational constraints, for example the restriction to only positive intensity values for incoherent photonic processors. In this work, we investigate the fundamental yet underexplored design and training challenges of physically constrained information processing with a particular focus on activation functions in non-negative neural networks (4Ns). Building on biological inspirations we revisit the concept of inhibitory (decreasing) and excitatory (increasing) activation functions, explore their effects experimentally and introduce a general approach for weight initialization of non-negative neural networks. Our results indicate the importance of both inhibitory and excitatory elements in activation functions in incoherent ONNs which should be considered for future design of optical activation functions for ONNs. Code is available at https://nnnn.cvmls.org | ||
| 650 | 4 | |a Activation functions | |
| 650 | 4 | |a Biomedical optical imaging | |
| 650 | 4 | |a constrained optimization | |
| 650 | 4 | |a deep learning | |
| 650 | 4 | |a Deep learning | |
| 650 | 4 | |a Network architecture | |
| 650 | 4 | |a Neural networks | |
| 650 | 4 | |a non-negative neural networks | |
| 650 | 4 | |a Optical computing | |
| 650 | 4 | |a Optical fiber networks | |
| 650 | 4 | |a optical neural networks | |
| 650 | 4 | |a Optical pulses | |
| 650 | 4 | |a Optical signal processing | |
| 650 | 4 | |a Photonics | |
| 650 | 4 | |a Training | |
| 700 | 1 | |a Drees, Dominik |e VerfasserIn |4 aut | |
| 700 | 1 | |a Brückerhoff-Plückelmann, Frank |d 1996- |e VerfasserIn |0 (DE-588)1336228814 |0 (DE-627)1895820448 |4 aut | |
| 700 | 1 | |a Schuck, Carsten |e VerfasserIn |4 aut | |
| 700 | 1 | |a Pernice, Wolfram |e VerfasserIn |0 (DE-588)103448429X |0 (DE-627)745631037 |0 (DE-576)382086015 |4 aut | |
| 700 | 1 | |a Risse, Benjamin |e VerfasserIn |4 aut | |
| 773 | 0 | 8 | |i Enthalten in |a Institute of Electrical and Electronics Engineers |t IEEE access |d New York, NY : IEEE, 2013 |g 13(2025), Seite 182474-182480 |h Online-Ressource |w (DE-627)728440385 |w (DE-600)2687964-5 |w (DE-576)373180713 |x 2169-3536 |7 nnas |
| 773 | 1 | 8 | |g volume:13 |g year:2025 |g pages:182474-182480 |g extent:7 |a Activation functions in non-negative neural networks |
| 856 | 4 | 0 | |u https://doi.org/10.1109/ACCESS.2025.3622408 |x Verlag |x Resolving-System |z kostenfrei |3 Volltext |7 0 |
| 856 | 4 | 0 | |u https://ieeexplore.ieee.org/document/11205509 |x Verlag |z kostenfrei |3 Volltext |7 0 |
| 951 | |a AR | ||
| 992 | |a 20260224 | ||
| 993 | |a Article | ||
| 994 | |a 2025 | ||
| 998 | |g 103448429X |a Pernice, Wolfram |m 103448429X:Pernice, Wolfram |d 130000 |d 130700 |e 130000PP103448429X |e 130700PP103448429X |k 0/130000/ |k 1/130000/130700/ |p 5 | ||
| 998 | |g 1336228814 |a Brückerhoff-Plückelmann, Frank |m 1336228814:Brückerhoff-Plückelmann, Frank |d 130000 |d 130700 |e 130000PB1336228814 |e 130700PB1336228814 |k 0/130000/ |k 1/130000/130700/ |p 3 | ||
| 999 | |a KXP-PPN1962407578 |e 4921791015 | ||
| BIB | |a Y | ||
| SER | |a journal | ||
| JSO | |a {"relHost":[{"type":{"bibl":"periodical","media":"Online-Ressource"},"physDesc":[{"extent":"Online-Ressource"}],"id":{"issn":["2169-3536"],"eki":["728440385"],"zdb":["2687964-5"]},"name":{"displayForm":["Institute of Electrical and Electronics Engineers"]},"corporate":[{"role":"aut","display":"Institute of Electrical and Electronics Engineers"}],"origin":[{"dateIssuedKey":"2013","publisher":"IEEE","dateIssuedDisp":"2013-","publisherPlace":"New York, NY"}],"pubHistory":["1.2013 -"],"recId":"728440385","title":[{"title":"IEEE access","subtitle":"practical research, open solutions","title_sort":"IEEE access"}],"note":["Gesehen am 24.10.12"],"language":["eng"],"disp":"Institute of Electrical and Electronics EngineersIEEE access","part":{"text":"13(2025), Seite 182474-182480","extent":"7","year":"2025","pages":"182474-182480","volume":"13"},"titleAlt":[{"title":"Access"}]}],"recId":"1962407578","title":[{"title_sort":"Activation functions in non-negative neural networks","title":"Activation functions in non-negative neural networks"}],"note":["Gesehen am 24.02.2026"],"language":["eng"],"person":[{"role":"aut","given":"Marlon","display":"Becker, Marlon","family":"Becker"},{"display":"Drees, Dominik","family":"Drees","role":"aut","given":"Dominik"},{"display":"Brückerhoff-Plückelmann, Frank","family":"Brückerhoff-Plückelmann","role":"aut","given":"Frank"},{"role":"aut","given":"Carsten","display":"Schuck, Carsten","family":"Schuck"},{"display":"Pernice, Wolfram","family":"Pernice","role":"aut","given":"Wolfram"},{"given":"Benjamin","role":"aut","family":"Risse","display":"Risse, Benjamin"}],"type":{"media":"Online-Ressource","bibl":"article-journal"},"physDesc":[{"extent":"7 S.","noteIll":"Illustrationen"}],"id":{"doi":["10.1109/ACCESS.2025.3622408"],"eki":["1962407578"]},"name":{"displayForm":["Marlon Becker, Dominik Drees, Frank Brückerhoff-Plückelmann, Carsten Schuck, Wolfram Pernice, and Benjamin Risse"]},"origin":[{"dateIssuedKey":"2025","dateIssuedDisp":"28 October 2025"}]} | ||
| SRT | |a BECKERMARLACTIVATION2820 | ||