solver.press

← All replicated papers

Did not replicate under our protocolNov 1, 2026

REVBRAIN: Real-time Probabilistic EEG Forecasting Using Residual Vector Quantization

Pankka, Lehtinen and Roine · arXiv 2610.38558

The outcome is the verdict on the one claim below, under the protocol we fixed in advance. It is not a verdict on the paper as a whole.

Claim tested

Median inference latency of autoregressive WaveNet is about 52x REVBRAIN's at 1 sample per input and about 3170x at 100 samples per input (released latency script).

What we ran and what came back

The authors' released latency script, with untrained weights, on a T4 GPU: 10 pairs of runs at each setting, 50 warm-up and 100 timed iterations. At one sample per input, WaveNet was 32 times slower than REVBRAIN (interval 29.9 to 34.0), which meets our bound of 26, set at half the paper's 52. At 100 samples per input it was 1,099 times slower (interval 1,019 to 1,185), which does not meet the bound of 1,585, half the paper's 3,170. The speed-up is large in both settings; at 100 samples it is about a third of the paper's figure on this hardware. REVBRAIN's own latency was similar to the paper's (about 1.4 to 2.3 milliseconds), while WaveNet took about 1,993 milliseconds here against 132 to 282 on the paper's larger GPUs, so the size of the ratio depends on how fast the baseline runs on a given card. Forecast accuracy, the real-time pipeline and the TF-REVBRAIN variant were not tested.

What this result supports

On a T4 with the released code, WaveNet is about 32x slower at 1 sample and about 1,100x slower at 100 samples.

What it does not support

The paper's 52x and 3170x figures; forecast accuracy; the laptop GPU; the TF-REVBRAIN variant; real-time pipeline latency.

Record

Protocol, deposited before the run: 10.5281/zenodo.23849261

The authors were sent this result. They have not reviewed this listing. The result and a report were sent on 1 November 2026. No reply had been received when this was listed.

How the protocol works is on the method page. If you are an author of this paper and something here is wrong, write to customers@aegismind.app and we will correct it on this page.