Research map / Probabilistic and scientific modeling / Bayesian optimization
Human-in-the-Loop Optimization for Deep Stimulus Encoding in Visual Prostheses
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use HILO to personalize the DSE using patient preference duels | DSE stimulus encoding · MNIST test, simulated patients | reconstruction error: DSE-φGuess: between .25 and 1.1; HILO: .071 ± .0031 | Helped |
| Use HILO to personalize the DSE using patient preference duels | DSE stimulus encoding · MNIST test, simulated patients | MNIST accuracy: DSE-φGuess: between 58.6% and 78.3%; HILO: 92% ± 1.0% | Helped |
| Replace the naive encoder with a DSE supplied with true patient parameters | Visual prosthesis stimulus encoding · MNIST test | perceptual loss: 5.68 -> 0.05 | Helped |
| Replace the naive encoder with a DSE supplied with true patient parameters | Visual prosthesis stimulus encoding · MNIST test | MNIST accuracy: 51% -> 95.6% | Helped |
| Use the proposed phosphene formulation instead of Granley et al. | Phosphene model · Electrode Location S1, S2, and S3 | shape evaluation score: S1: 2.43 -> 0.28; S2: 7.07 -> 0.57; S3: 1.15 -> 0.38 | Helped |
| Use the proposed phosphene formulation instead of Granley et al. | Phosphene model · Stimulus Parameters 4A, 4B, 4C, and 5 | mean squared error: 4A: 0.9 -> 0.73; 4B: 2.1 -> 2.3; 4C: 0.16 -> 0.1; 5: 49.5 -> 48.6 | Mixed |
| Increase simulated patient feedback noise to σ=0.05 (ablation) | HILO · MNIST test, simulated patients | reconstruction error | Mixed |
| Use a misspecified axon trajectory in the encoder's forward model (ablation) | HILO · Simulated patients with varied axon trajectories | reconstruction error | No clear effect |
| Use incorrect stimulation thresholds in the encoder's forward model (ablation) | HILO · Simulated patients with threshold misspecification | reconstruction error | Hurt |
| Optimize HILO for patient parameters outside the encoder's training ranges (ablation) | HILO · Out-of-distribution simulated patients | reconstruction error | Mixed |
| Use Matérn 5/2 rather than the other tested GP kernels (ablation) | HILO preference model · 20 simulated patients | reconstruction error | Helped |
| Use patient-optimal rather than transfer-selected GP hyperparameters (ablation) | HILO preference model · 20 simulated patients | final reconstruction error: p>.05 | No clear effect |
Closest papers by meaning
- RTify: Aligning Deep Neural Networks with Human Behavioral Decisions (NeurIPS 2024 · similarity 0.82)
- Adapting Pretrained ViTs with Convolution Injector for Visuo-Motor Control (ICML 2024 · similarity 0.82)
- Model–Behavior Alignment under Flexible Evaluation: When the Best-Fitting Model Isn’t the Right One (NeurIPS 2025 · similarity 0.82)
- FOCUS: Benchmarking Retinal Model Generalization from Foundation Vision Encoders to Multimodal LLMs (NeurIPS 2026 · similarity 0.82)
- NEvo: Neural-Guided Evolutionary Video Synthesis for Dynamic Visual Selectivity (NeurIPS 2026 · similarity 0.82)
- BaRISTA: Brain Scale Informed Spatiotemporal Representation of Human Intracranial Neural Activity (NeurIPS 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.