On the biological plausibility of orthogonal initialisation for solving gradient instability in deep neural networks
This addresses a biological plausibility problem for neuroscience and AI researchers, but is incremental as it builds on known orthogonal initialization methods.
The paper tackled the biological plausibility issue of orthogonal initialization in neural networks by proposing two schemes that allow networks to evolve orthogonal weights naturally, showing they outperform random initialization in recurrent and feedforward networks.
Initialising the synaptic weights of artificial neural networks (ANNs) with orthogonal matrices is known to alleviate vanishing and exploding gradient problems. A major objection against such initialisation schemes is that they are deemed biologically implausible as they mandate factorization techniques that are difficult to attribute to a neurobiological process. This paper presents two initialisation schemes that allow a network to naturally evolve its weights to form orthogonal matrices, provides theoretical analysis that pre-training orthogonalisation always converges, and empirically confirms that the proposed schemes outperform randomly initialised recurrent and feedforward networks.