← Return to the planispherePlate VII · Newcastle / Ramaswamy · O2 lead

PLATE VII · MMXXVI · NEWCASTLE / RAMASWAMYA column, four modulators, twenty yearsMATHEMATICAL PRIMARY · MEI · MULLER · RAMASWAMY 2022LAYER ImolecularLAYER II / IIIpyramidalLAYER IVgranularLAYER VpyramidalLAYER VImultiformWHITE MATTERaxonalAchACETYLCHOLINEDADOPAMINENENORADRENALINE5-HTSEROTONINMODULATOR · ACHAcetylcholine
Attention and uncertainty. Modulates the gain of cortical processing under attentional demand — the model's primary substrate for surfacing salient sub-networks.
Newcastle University · School of Computing

Sri Ramaswamy

Chair of computational neuroscience. Co-author of the four-scale neuromodulatory framework SYMPHONY transposes from cortex to code. Lead of Objective O2 and co-lead of the ethics layer of O5.


I · Position

From Blue Brain to the chair at Newcastle

Sri Ramaswamy is chair of computational neuroscience at Newcastle University’s School of Computing. His twenty-year arc moves from the digital reconstruction of cortical microcircuits at the Blue Brain Project at EPFL through an independent group on biologically-grounded artificial neural networks to the foundational framework that gives SYMPHONY its mathematical primary source: the four-scale neuromodulatory formulation in Trends in Neurosciences, 2022.

The pedigree is the moat. Mei, Muller & Ramaswamy (2022) is not a paper SYMPHONY references politely from a distance — it is the mathematical scaffolding on which O2 is built, with the third author as the consortium’s objective lead.

II · The 2022 paper

Mei · Muller · Ramaswamy — four scales, one substrate

The 2022 paper formalised a four-scale framework for integrating neuromodulation into deep networks — hyperparameter scale (global gain), plasticity scale (connectivity reshape), neuronal scale (per-node gating), and dendritic scale (branch-local computation). The simulation evidence showed three behavioural properties that SYMPHONY requires from its substrate.

Property iMei et al. 2022

Faster adaptation

New tasks reach competence in fewer training steps than a non-modulated baseline.

Property iiMei et al. 2022

Higher cumulative reward

Across task sequences the modulated network accumulates more reward than the static one.

Property iiiMei et al. 2022

Resistance to catastrophic forgetting

The same network retains old skills as new ones land — the prerequisite for a substrate that survives an engineer's day.

SYMPHONY’s critical uncertainty is whether these three properties survive transposition from a continuous perceptual domain to a discrete symbolic one. That question is what O2 resolves.

III · Pedigree

Twenty years from cortex to code

L1L6CORTICAL COLUMN · L1–L6AchDANE5-HTNEUROMODULATOR BEAMS200520262005Blue Brain begins2008PhD in computational neuroscience2015Cortical-column reconstruction2018Independent group2022Mei · Muller · Ramaswamy2024Newcastle chair2026SYMPHONY · O2 lead
Two decades of biologically-grounded neuroscience, ending in SYMPHONY's O2. Hover any beat for the source and the role it plays in the proposal. Primary beats — Cell 2015, TINS 2022 — are the load-bearing citations.
IV · Role in SYMPHONY

O2 lead · co-lead of the ethics layer of O5

Objective O2M18 decision milestone

Implement, on the substrate produced by O1, a computational instantiation of the four-scale framework of Mei, Muller & Ramaswamy (2022), adapted from continuous perceptual signals to discrete symbolic activations. Threshold: a single trained substrate demonstrates statistically significant (p < 0.01, paired test, ≥ 30 task instances) task-appropriate subnetwork activation across at least three distinct engineering task classes (localisation, impact analysis, refactoring candidate discovery), with F1 ≥ 0.6 versus expert-annotated relevance.

Newcastle co-leads the ethics layer of O5 — the equitable-access user study — bringing the school’s established protocol for stratified pre-registered behavioural studies in computing education.