Towards Stream Learning on Embedded Systems: Benchmarking the Memory Consumption of Stream Learning Methods
Sebastian Buschjäger, Nuwan Gunasekara, Heitor Murilo Gomes · Read the paper
Hook: A stream learner can maintain predictive performance yet still become unusable when its memory footprint exceeds an embedded system’s budget.
Summary: The paper benchmarks seven stream classifiers across 13 real and synthetic streams, using model-size budgets from 128 KiB to approximately 8 MiB and 6,463 experiments. It measures failure-aware accuracy, peak model size, time to budget exhaustion, and prediction-plus-update latency to assess sustained operation on resource-constrained systems.
Industry impact: The reported results identify two resource failure modes: adaptive ensembles may exceed small budgets immediately, while incremental trees can grow substantially during long streams. The paper finds that explicitly compact methods are generally the only viable choices under the smallest budgets, whereas adaptive ensembles become competitive as more memory is available.
Potential implications: The authors conclude that many state-of-the-art stream-learning methods are only partially applicable to embedded or long-running systems. They call for bounded resource usage to become a first-class design objective and propose an API that lets stream learners expose and respect resource budgets.
Conjoint Audio-to-Spikes Encoding and Processing for Efficient Neuromorphic Speech Recognition
Valentin M. Meunier, Amélie Gruel, Pierre Lewden, Adrien F. Vincent, Sylvain Saïghi · Read the paper
Hook: The study treats the encoder and classifier as a joint design problem, aiming to balance informative spike representations, accuracy, and spiking activity.
Summary: The paper evaluates a non-learnable, programmable audio-to-spikes encoder designed for hardware implementation on FPGAs, together with a classifier for neuromorphic speech recognition. The authors report an end-to-end spike encoding and evaluation of TIMIT and a 99.77% classification accuracy on spike-encoded Heidelberg Digits.
Industry impact: For teams exploring neuromorphic speech processing, the work provides an example of a software encoding pipeline explicitly targeting digital hardware implementation. The paper evaluates efficiency using hardware-agnostic metrics based on quantitative spiking activity rather than relying only on classification accuracy.
Potential implications: The reported results suggest that encoder design can affect both recognition performance and the energy cost of learning and inference in spiking systems. Further work would be needed to determine how these methods perform on deployed FPGA systems and beyond the datasets evaluated in the paper.
