UCogNet
Technical noteRequest a demo

Research visualizations

Auto-aggregated figure index across every research domain. New figures appear here as soon as their manifest.json is added to public/. Click any chart to view at full resolution.

93 figures · 6 domainsindex generated 2026-05-30
Latest additionsnewest figures across all domains
v0.2 audit — full 11×11 confusion matrix (row-normalised)
Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)
2026-05-25

v0.2 audit — full 11×11 confusion matrix (row-normalised)

v0.2 audit — per-class precision / recall / F1 (with macro and weighted F1)
Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)
2026-05-25

v0.2 audit — per-class precision / recall / F1 (with macro and weighted F1)

v0.2 audit — overall accuracy and per-class recall with 95 % confidence intervals
Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)
2026-05-25

v0.2 audit — overall accuracy and per-class recall with 95 % confidence intervals

v0.2 audit — coverage–accuracy abstention curve
Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)
2026-05-25

v0.2 audit — coverage–accuracy abstention curve

v0.2 audit — ResNet18 embedding nearest-neighbour leakage check (test → train)
Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)
2026-05-25

v0.2 audit — ResNet18 embedding nearest-neighbour leakage check (test → train)

v0.2 audit — perceptual-hash leakage check (test → train)
Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)
2026-05-25

v0.2 audit — perceptual-hash leakage check (test → train)

Veterinary microscopy AI — parasite egg classification, 11 species (v0.2 audit)

ResNet18 + 45-module integrated system + Tier-3 safety stack · 97.36 percent (95% CI [96.24, 98.16])
17 fig · 2026-05-25Full page →
v0.2 audit — full 11×11 confusion matrix (row-normalised)

v0.2 audit — full 11×11 confusion matrix (row-normalised)

🔍
v0.2 audit — per-class precision / recall / F1 (with macro and weighted F1)

v0.2 audit — per-class precision / recall / F1 (with macro and weighted F1)

🔍
v0.2 audit — overall accuracy and per-class recall with 95 % confidence intervals

v0.2 audit — overall accuracy and per-class recall with 95 % confidence intervals

🔍
v0.2 audit — coverage–accuracy abstention curve

v0.2 audit — coverage–accuracy abstention curve

🔍
v0.2 audit — ResNet18 embedding nearest-neighbour leakage check (test → train)

v0.2 audit — ResNet18 embedding nearest-neighbour leakage check (test → train)

🔍
v0.2 audit — perceptual-hash leakage check (test → train)

v0.2 audit — perceptual-hash leakage check (test → train)

🔍
v0.2 robustness — safety-first scatter (Δaccuracy vs Δconfidence)

v0.2 robustness — safety-first scatter (Δaccuracy vs Δconfidence)

🔍
v0.2 robustness — accuracy and ECE under each perturbation / severity

v0.2 robustness — accuracy and ECE under each perturbation / severity

🔍
v0.2 robustness — abstention rate at τ = 0.90 under each perturbation

v0.2 robustness — abstention rate at τ = 0.90 under each perturbation

🔍
Original headline figure (v0.1 framing) — 97.4 % on 11-class Chula

Original headline figure (v0.1 framing) — 97.4 % on 11-class Chula

🔍
Detection gallery — real micrographs

Detection gallery — real micrographs

🔍
Grad-CAM attention — how the model sees

Grad-CAM attention — how the model sees

🔍
Confusion matrix — v0.1 visualisation

Confusion matrix — v0.1 visualisation

🔍
Per-class recall — two channels

Per-class recall — two channels

🔍
Calibration — per-class temperature scaling

Calibration — per-class temperature scaling

🔍
Training dynamics

Training dynamics

🔍
Conv-1 filter bank

Conv-1 filter bank

🔍

Biomolecular safety screening (toxin pre-flight check)

Post-humanist 5-axis perception + Deep MLP + AnankeProtocol over 16,466 peptides (ToxinPred-2)
16 fig · 2026-05-24Full page →
Training dynamics — Deep MLP head on the 5-axis perception

Training dynamics — Deep MLP head on the 5-axis perception

🔍
Per-axis ablation — which modality carries the discriminative signal

Per-axis ablation — which modality carries the discriminative signal

🔍
Accuracy stratified by sequence-length quintile

Accuracy stratified by sequence-length quintile

🔍
Principal-component projection of the 64-D perception

Principal-component projection of the 64-D perception

🔍
Cognitive-module signals across training

Cognitive-module signals across training

🔍
Metacognition L1 — Ψ-vector across training

Metacognition L1 — Ψ-vector across training

🔍
Ananke Protocol — safety verdict timeline

Ananke Protocol — safety verdict timeline

🔍
Reliability diagram and per-class confidence distribution

Reliability diagram and per-class confidence distribution

🔍
Mean Kyte–Doolittle hydrophobicity profile per class

Mean Kyte–Doolittle hydrophobicity profile per class

🔍
Spectral fingerprint of the hydrophobicity track

Spectral fingerprint of the hydrophobicity track

🔍
Persistent-homology surrogate — Betti curves per class

Persistent-homology surrogate — Betti curves per class

🔍
Synaesthetic coherence — per-class cross-axis Pearson matrices

Synaesthetic coherence — per-class cross-axis Pearson matrices

🔍
Hardest misclassifications — high-confidence errors in PCA space

Hardest misclassifications — high-confidence errors in PCA space

🔍
Cognitive-module signal synergy across training checkpoints

Cognitive-module signal synergy across training checkpoints

🔍
Sequence-length distributions per split

Sequence-length distributions per split

🔍
Per-amino-acid composition shifts (toxic − non-toxic)

Per-amino-acid composition shifts (toxic − non-toxic)

🔍

Astrophysical anomaly detection — Cambioides on Planck SMICA

C3-symmetric Thomas attractor as a third null model · Level-A calibrated
4 fig · 2026-05-24Full page →
Mollweide of −log10 p

Mollweide of −log10 p

🔍
p-value distribution + KS test

p-value distribution + KS test

🔍
Significance vs Galactic latitude

Significance vs Galactic latitude

🔍
Q–Q against Uniform(0,1)

Q–Q against Uniform(0,1)

🔍

Patient-aware cardio arrhythmia detection (MIT-BIH AAMI inter-patient)

Tier-3 safety stack on top of Tier-2 weights · no retraining · F-rescue x8 via free-energy
16 fig · 2026-05-14Full page →
Tier 3 headline — V preserved, ECE −33%, F-rescue ×8

Tier 3 headline — V preserved, ECE −33%, F-rescue ×8

🔍
Coverage vs accuracy (selective prediction)

Coverage vs accuracy (selective prediction)

🔍
Free-energy AUROC per arrhythmia class

Free-energy AUROC per arrhythmia class

🔍
F-rescue ROC — moat finding

F-rescue ROC — moat finding

🔍
L2 class-aware drift events on DS2 stream

L2 class-aware drift events on DS2 stream

🔍
Tier 2 headline — 73.6% / F1=0.358 on full DS2

Tier 2 headline — 73.6% / F1=0.358 on full DS2

🔍
Tier 2 per-class recall on full DS2

Tier 2 per-class recall on full DS2

🔍
Tier 1 → Tier 2 ablation

Tier 1 → Tier 2 ablation

🔍
Inter-patient AAMI vs literature

Inter-patient AAMI vs literature

🔍
Tier 2 training dynamics

Tier 2 training dynamics

🔍
Cardio Tier 1 headline (legacy)

Cardio Tier 1 headline (legacy)

🔍
Cardio Tier 1 per-class (legacy)

Cardio Tier 1 per-class (legacy)

🔍
Cardio Tier 1 training (legacy)

Cardio Tier 1 training (legacy)

🔍
Cardio Ψ-trajectory (cognitive state)

Cardio Ψ-trajectory (cognitive state)

🔍
Cardio cognitive module signals

Cardio cognitive module signals

🔍
Cross-domain comparison

Cross-domain comparison

🔍

Post-humanist parasite vision — autonomous self-detection of classification ceiling

Chula-ParasiteEgg-11 · 5 modal axes · zero learned convolutions · metacognition L2 fires unsupervised
5 fig · 2026-05-08Full page →
Headline accuracy — random 9.1% · cosine 17.8% · trained head 26.7%

Headline accuracy — random 9.1% · cosine 17.8% · trained head 26.7%

🔍
Self-consistency trajectory

Self-consistency trajectory

🔍
Cognitive module signals — three concur

Cognitive module signals — three concur

🔍
Training dynamics

Training dynamics

🔍
Per-class recall

Per-class recall

🔍

Platform diagnostics & plasma turbulence control benchmarks

Hasegawa–Wakatani 2D · 10-module cognitive pyramid · 5 OOD evaluation campaigns · 6 seeds
35 fig · 2026-04-01Full page →
Plasma — UCogNet vs HW baselines

Plasma — UCogNet vs HW baselines

🔍
Plasma — energy trajectories

Plasma — energy trajectories

🔍
Plasma — budget vs performance Pareto

Plasma — budget vs performance Pareto

🔍
Plasma — out-of-distribution gap

Plasma — out-of-distribution gap

🔍
Plasma — ablation study

Plasma — ablation study

🔍
Module capability radar

Module capability radar

🔍
Campaign heatmap

Campaign heatmap

🔍
Regime dynamics

Regime dynamics

🔍
Hypothesis distribution

Hypothesis distribution

🔍
Module symbiosis matrix

Module symbiosis matrix

🔍
Confidence phase plot

Confidence phase plot

🔍
Dashboard summary

Dashboard summary

🔍
Test coverage

Test coverage

🔍
Regime radar

Regime radar

🔍
Confidence signals

Confidence signals

🔍
Routing heatmap

Routing heatmap

🔍
Reward composition

Reward composition

🔍
Cost breakdown

Cost breakdown

🔍
Evidence asymmetry

Evidence asymmetry

🔍
Mutation intensity

Mutation intensity

🔍
Rollout pipeline

Rollout pipeline

🔍
Statistical arsenal

Statistical arsenal

🔍
Architecture pyramid

Architecture pyramid

🔍
Shaping guardrails

Shaping guardrails

🔍
Campaign results

Campaign results

🔍
Integration flow

Integration flow

🔍
Capability boundaries

Capability boundaries

🔍
Mode ablation

Mode ablation

🔍
Parameter efficiency

Parameter efficiency

🔍
Cost vs performance

Cost vs performance

🔍
Combined leaderboard

Combined leaderboard

🔍
Leaderboard — ARC-AGI-2

Leaderboard — ARC-AGI-2

🔍
Leaderboard — HLE rolling

Leaderboard — HLE rolling

🔍
Leaderboard — LiveBench

Leaderboard — LiveBench

🔍
Leaderboard — SWE-Bench Lite

Leaderboard — SWE-Bench Lite

🔍