VIXLSTM● GENERATED
2021 · ACM · Research

VixLSTM.

The model got the prediction right — but can you see why it believed it?

Stack
LSTMExplainable AIPyTorchD3.js

LSTMs are formidable at modeling sequences, and almost completely opaque about how they do it. VixLSTM is a visual interface that peels back the recurrence — showing which inputs, at which time steps, pushed the network toward its prediction.

VIXLSTM● GENERATED
A grid of recurrent cells with a saliency overlay: warmer cells contributed more to the output. Reading left-to-right traces how influence accumulates across the sequence.

Explainability you can interrogate

Rather than a single static attribution score, VixLSTM lets you brush a time range, isolate a feature, and watch the model’s internal attention respond. The goal is not just post-hoc justification but a working understanding analysts can trust — and challenge.

A prediction you cannot explain is a prediction you cannot deploy responsibly.

Built with collaborators and published through ACM in 2021, VixLSTM sits at the meeting point of machine learning and visual analytics — making a model’s reasoning as inspectable as its output.