Although deep feedforward neural networks share some characteristics with the primate visual system, a key distinction is their dynamics. Deep nets typically operate in serial stages wherein each layer completes its computation before processing begins in subsequent layers. In contrast, biological systems have cascaded dynamics: information propagates from neurons at all layers in parallel but transmission occurs gradually over time, leading to speed-accuracy trade offs even in feedforward architectures. We explore the consequences of biologically inspired parallel hardware by constructing cascaded ResNets in which each residual block has propagation delays but all blocks update in parallel in a stateful manner. Because information transmitted through skip connections avoids delays, the functional depth of the architecture increases over time, yielding anytime predictions that improve with internal-processing time. We introduce a temporal-difference training loss that achieves a strictly superior speed-accuracy profile over standard losses and enables the cascaded architecture to outperform state-of-the-art anytime-prediction methods. The cascaded architecture has intriguing properties, including: it classifies typical instances more rapidly than atypical instances; it is more robust to both persistent and transient noise than is a conventional ResNet; and its time-varying output trace provides a signal that can be exploited to improve information processing and inference.

Although deep feedforward neural networks share some characteristics with the primate visual system, a key distinction is their dynamics. Deep nets typically operate in serial stages wherein each layer completes its computation before processing begins in subsequent layers. In contrast, biological systems have cascaded dynamics: information propagates from neurons at all layers in parallel but transmission occurs gradually over time, leading to speed-accuracy trade offs even in feedforward architectures. We explore the consequences of biologically inspired parallel hardware by constructing cascaded ResNets in which each residual block has propagation delays but all blocks update in parallel in a stateful manner. Because information transmitted through skip connections avoids delays, the functional depth of the architecture increases over time, yielding anytime predictions that improve with internal-processing time. We introduce a temporal-difference training loss that achieves a strictly superior speed-accuracy profile over standard losses and enables the cascaded architecture to outperform state-of-the-art anytime-prediction methods.

Although deep feedforward neural networks share some characteristics with the primate visual system, a key distinction is their dynamics. Deep nets typically operate in serial stages wherein each layer completes its computation before processing begins in subsequent layers. In contrast, biological systems have cascaded dynamics: information propagates from neurons at all layers in parallel but transmission occurs gradually over time, leading to speed-accuracy trade offs even in feedforward architectures. We explore the consequences of biologically inspired parallel hardware by constructing cascaded ResNets in which each residual block has propagation delays but all blocks update in parallel in a stateful manner. Because information transmitted through skip connections avoids delays, the functional depth of the architecture increases over time, yielding anytime predictions that improve with internal-processing time. We introduce a temporal-difference training loss that achieves a strictly superior speed-accuracy profile over standard losses and enables the cascaded architecture to outperform state-of-the-art anytime-prediction methods.

Neural networks today often recognize objects as well as people do, and thus might serve as models of the human recognition process. However, most such networks provide their answer after a fixed computational effort, whereas human reaction time varies, e.g. from 0.2 to 10 s, depending on the properties of stimulus and task. To model the effect of difficulty on human reaction time, we considered a classification network that uses early-exit classifiers to make anytime predictions. Comparing human and MSDNet accuracy in classifying CIFAR-10 images in added Gaussian noise, we find that the network equivalent input noise SD is 15 times higher than human, and that human efficiency is only 0.6\% that of the network. When appropriate amounts of noise are present to bring the two observers (human and network) into the same accuracy range, they show very similar dependence on duration or FLOPS, i.e. very similar speed-accuracy tradeoff. We conclude that Anytime classification (i.e. early exits) is a promising model for human reaction time in recognition tasks.

The pioneer deep neural networks (DNNs) have emerged to be deeper or wider for improving their accuracy in various applications of artificial intelligence. However, DNNs are often too heavy to deploy in practice, and it is often required to control their architectures dynamically given computing resource budget, i.e., anytime prediction. While most existing approaches have focused on training multiple shallow sub-networks jointly, we study training thin sub-networks instead. To this end, we first build many inclusive thin sub-networks (of the same depth) under a minor modification of existing multi-branch DNNs, and found that they can significantly outperform the state-of-art dense architecture for anytime prediction. This is remarkable due to their simplicity and effectiveness, but training many thin sub-networks jointly faces a new challenge on training complexity. To address the issue, we also propose a novel DNN architecture by forcing a certain sparsity pattern on multi-branch network parameters, making them train efficiently for the purpose of anytime prediction. In our experiments on the ImageNet dataset, its sub-networks have up to $43.3\%$ smaller sizes (FLOPs) compared to those of the state-of-art anytime model with respect to the same accuracy. Finally, we also propose an alternative task under the proposed architecture using a hierarchical taxonomy, which brings a new angle for anytime prediction.