Biologically grounded neocortex computational primitives implemented on neuromorphic hardware improve vision transformer performance.
In vitro record. Proceedings of the National Academy of Sciences of the United States of America, 2025.
- Cell
- 2025
Authorship
Exact sentences from the abstract. No machine-written text is on this page. No person reviewed it.
Source: the abstract of PMID 41055996. Checked Verification: Evidence verified by machine against declared source
This study and its findings belong to the authors cited. Defiance International, a supplier of chemical sequences for research use, keeps this record and makes no claim from its findings.
Evidence verified by machine against declared source
Key facts
- Design
- Not reported
- System
- Cell
- Scale
- It works as follows: The input image is distributed into N (flattened 2D) patches (where we keep N = 6 for the MNIST experiments), and linear embeddings of these patches are fed as input to the encoder of the trained ViT.
- Comparator
- None reported
- Journal
- Proceedings of the National Academy of Sciences of the United States of America
- Year
- 2025
- PMID
- 41055996
- Verification
- Evidence verified by machine against declared source The stored extraction was checked against its declared source. Public summary availability is separate. No tier is an endorsement, an evidence-strength rating, or independent replication, and none establishes efficacy, safety, or suitability.
- Full-text availability
- Abstract only
Technical summary
Study scale
It works as follows: The input image is distributed into N (flattened 2D) patches (where we keep N = 6 for the MNIST experiments), and linear embeddings of these patches are fed as input to the encoder of the trained ViT.
Outcomes examined
By parametrically varying synaptic conductances within physiologically realistic bounds, we quantified the distinct computational roles of each inhibitory subtype in shaping pyramidal output and circuit-level dynamics (Fig. 2 D and E).
Key findings
Hardware implementations demonstrated linear compute-time scaling with load, contrasting the quadratic scaling of simulations.
Not reported in the declared source: Methods, Study design, Participants or experimental system, Intervention or exposure, Comparator. Not published in this record: Research question.
Limitations and uncertainty
Not reported in the declared source.
Cite
Cite the study
Biologically grounded neocortex computational primitives implemented on neuromorphic hardware improve vision transformer performance.
Iqbal A, Mahmood H, Stuart GJ et al. Proceedings of the National Academy of Sciences of the United States of America. 2025. PMID 41055996
The study and its abstract belong to their authors and publisher. Cite them for anything the study found.
Report an error
If a sentence on this record does not match the abstract, tell us which one. A report starts an automatic re-check by machine. No person reads the record first.
The form takes a sentence and your note. It asks for no name and no address, and we store neither. Do not put personal details in the note.
The report form runs in the interactive page. Open this record with scripts enabled to send a report.
Corrections on this record
No correction has been recorded for this record.
Method sheet
No machine-drafted summary is published on this record. Any section shown is the abstract’s own sentence, printed as issued.