Behaviourally relevant sensory input is routinely transmitted and processed in biological networks, despite perpetual perceptually unrelated external, and spontaneously generated internal activity. Such activity has traditionally been understood as noise. In feed-forward networks, such noise can pose a challenge, because disambiguating stimulus-evoked activity and noise in deeper layers becomes increasingly difficult, for example in pattern classification tasks. On the other hand, the beneficial effects of noise for stochastic regularization and preventing co-adaptation were recognized in deep-learning, but it remains unclear how to weigh the pros and cons. Biological networks may help to shed light on how both regularization and resilience to the detrimental effects of noise can be implemented locally. Recent results link the spontaneous activity of individual neurons to their metabolic homeostasis, making it computationally intertwined with signal processing. In particular, neural metabolism affects a neuron’s response function, by providing constraints on resource availability. Such metabolic constraints of individual neurons in feed-forward networks could serve as a local regularization mechanism against input noise. Here, we investigate how a functional form of neuronal metabolism based on resource consumption influences a neuron’s response curve, and global activation patterns in a feed-forward network, both in the absence and presence of input stimuli and synaptic noise. Further, we show that introducing noise that deteriorates network performance in the classification task can be effectively counteracted locally by considering resource limitations of individual neurons, leading to an increased resilience of signal processing.