Estimating cognitive workload from physiological signals is becoming increasingly important in high‑risk settings, where continuous and unobtrusive monitoring can inform adaptive systems and safety interventions. This paper introduces FOCUS (Frequency-band Optimized Cognitive-load Ultralight SSM), a compact model built on Mamba‑2 for binary EEG workload classification under limited data conditions. Instead of learning frequency sensitivity, FOCUS encodes pre‑processed multi‑channel EEG with a fixed Fast Fourier Transform (FFT) that decomposes the signal into the five canonical bands (δ, θ, α, β, γ). The architecture combines depthwise‑separable convolution with a weight‑shared encoder‑bottleneck Mamba‑2 pair, totaling 0.04 M parameters and 0.11 GFLOPs.
In a pilot study with 11 participants, leave‑one‑subject‑out cross‑validation was applied across three cognitively distinct tasks. FOCUS achieved 74.75% accuracy on the full dataset with a GPU forward‑pass latency of only 0.08 ms. An ablation on weight sharing showed that the shared configuration improved accuracy by 4.55 percentage points compared to the non‑shared variant while reducing the SSM parameter budget; however, the difference was not statistically significant under a two‑sided Wilcoxon signed‑rank test ($p=0.083$).
Overall, the selected FOCUS configuration offers a favorable trade‑off between classification performance and computational cost in this controlled pilot setting, making it suitable for real‑time workload monitoring on resource‑constrained devices. Review