# Evaluating Deep Human-in-the-Loop Optimization for Retinal Implants Using Sighted Participants

arXiv: https://arxiv.org/abs/2502.00177 · HTML: https://mlsanity.com/paper/2502.00177/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 8 of its 42 references that are in the map.
- First posted on arXiv 31 Jan 2025.

## Builds on (most cited first, 8 of 8)

- [PieAPP: Perceptual Image-Error Assessment through Pairwise Preference](https://mlsanity.com/paper/1806.02067/) (2018)
- [Hybrid Neural Autoencoders for Stimulus Encoding in Visual and Other Sensory Neuroprostheses](https://mlsanity.com/paper/2205.13623/) (2022)
- [Human-in-the-Loop Optimization for Deep Stimulus Encoding in Visual Prostheses](https://mlsanity.com/paper/2306.13104/) (2023)
- [Deep Learning-Based Perceptual Stimulus Encoder for Bionic Vision](https://mlsanity.com/paper/2203.05604/) (2022)
- [Brain Co-Processors: Using AI to Restore and Augment Brain Function](https://mlsanity.com/paper/2012.03378/) (2020)
- [Adapting Brain-Like Neural Networks for Modeling Cortical Visual Prostheses](https://mlsanity.com/paper/2209.13561/) (2022)
- [Assistive technology use in domestic activities by people who are blind](https://mlsanity.com/paper/2305.03019/) (2023)
- [Beyond Sight: Probing Alignment Between Image Models and Blind V1](https://mlsanity.com/paper/2403.12990/) (2024)

## Built on by (most cited first, 1 of 1)

- [Fuzzing the brain: Automated stress testing for the safety of ML-driven neurostimulation](https://mlsanity.com/paper/2512.05383/) (2025)

## Abstract (arXiv)

Human-in-the-loop optimization (HILO) is a promising approach for personalizing visual prostheses by iteratively refining stimulus parameters based on user feedback. Previous work demonstrated HILO's efficacy in simulation, but its performance with human participants remains untested. Here we evaluate HILO using sighted participants viewing simulated prosthetic vision to assess its ability to optimize stimulation strategies under realistic conditions. Participants selected between phosphenes generated by competing encoders to iteratively refine a deep stimulus encoder (DSE). We tested HILO in three conditions: standard optimization, threshold misspecifications, and out-of-distribution parameter sampling. Participants consistently preferred HILO-generated stimuli over both a naive encoder and the DSE alone, with log odds favoring HILO across all conditions. We also observed key differences between human and simulated decision-making, highlighting the importance of validating optimization strategies with human participants. These findings support HILO as a viable approach for adapting visual prostheses to individuals. Clinical relevance: Validating HILO with sighted participants viewing simulated prosthetic vision is an important step toward personalized calibration of future visual prostheses.