← Back to the paper gallery

Paper story

Teaching a network to fill in missing signals

The Weighted Gate Layer Autoencoder, explained without the mathematics: why we deliberately blindfold a neural network during training, and how that lets it redraw sensor channels that die mid-recording.

Paper
Weighted Gate Layer Autoencoders
Venue
IEEE Transactions on Cybernetics
Year
2021
DOI
10.1109/TCYB.2021.3049583
Code & data
IEEE Code Ocean capsule

The problem

Sensors fail at the worst possible time

Many measurements worth taking are taken many channels at once. An EEG headset records dozens of electrodes; an aircraft, a factory line, or a robot fuses readings from banks of sensors. And in every one of these settings the same small disaster is routine: mid-recording, a channel dies. An electrode loses contact, a wire is knocked, a sensor drifts out of range.

The standard responses are painful. Throw the whole recording away, and hours of experiment time go with it. Interpolate the missing channel from its neighbours with a fixed formula, and you get a guess that ignores everything the data could have told you about how that specific channel actually behaves.

The idea

Practise losing, so losing doesn't hurt

An autoencoder is a neural network trained to reproduce its own input: it squeezes the data through a narrow middle and expands it back, and in doing so it is forced to learn the structure that makes the data compressible. Our addition is a gate layer in front of it. During training, gates close at random, switching individual input channels off — but the network is still asked to reconstruct all channels at the output, including the ones it never saw.

THE GATE LAYER inputs · gate layer x1x2x3x4x5 encoder decoder outputs x3 reconstructedwithout being seen gates close at random in training → the network must infer the hidden input from the rest

One gate closed: the network receives four of the five inputs, yet must still produce all five outputs. The only way to succeed is to learn how the channels depend on one another.

A weighting mechanism sharpens the lesson: the reconstruction error on the switched-off channels counts for more, so training concentrates on exactly the skill that matters — recovering what is missing, not merely copying what is present. The result is a model that has spent its entire training life practising sensor failure, long before it meets a real one.

What we found

Graceful degradation instead of collapse

At test time, the difference shows up exactly where it should. When a channel drops out, the gated model redraws it from the surviving channels — not as a smoothed average, but as a signal that tracks the true one. And as more channels are removed, its reconstruction error stays nearly flat where a standard autoencoder’s error climbs steeply: with ten channels missing at once, the standard model has effectively given up while the gated one is still working.

FILLING IN A LOST CHANNEL recorded channels channel 3 drops out mid-recording gated autoencoder reconstructed redrawn from the other channels alone reconstruction vs true signal trained with gates randomly closed, the model learns cross-channel structure ERROR VS CHANNELS LOST standard AE gated AE 1 10 missing

Left: a channel dies mid-recording. Right: the gated autoencoder redraws it from the other channels alone; the inset shows error staying nearly flat as up to ten channels are removed.

Who should care

Anywhere data arrives on many channels

We developed this with EEG in mind — brain-computer interfaces are notorious for electrodes dropping out mid-session — but the mechanism is domain-agnostic. Wearables, industrial monitoring, autonomous platforms: any system that fuses multiple sensors inherits the same failure mode, and the same fix applies. The gate layer also has a second life as an analysis tool: because the model learns which channels can stand in for which, it exposes the dependency structure of the data itself.

This paper seeded a line of work: the original gate-layer idea appeared at IJCNN 2019, and the follow-up in Sensors (2025), led by Kangjing Li, a postdoctoral researcher who worked on my project under my supervision, asks how the order in which channels are gated should follow the brain’s own layout.

Cite & explore

The formal version

H. El-Fiqi, M. Wang, K. Kasmarik, A. Bezerianos, K. C. Tan and H. A. Abbass, “Weighted Gate Layer Autoencoders,” IEEE Transactions on Cybernetics, vol. 52, no. 8, pp. 7242–7253, 2022. doi:10.1109/TCYB.2021.3049583

Run the code and data on Code Ocean →
Back to the paper gallery →

How this page was written. The research, the results, and the ideas here are mine and my co-authors’. To retell them in plain language, I worked with an AI writing assistant that helped draft the text and render the diagrams in this site’s style. I reviewed and edited everything, and the technical responsibility rests with me. If the prose reads a little differently from my papers, that is why.