No abstract

No abstract

Thanks to the rapid proliferation of connected devices, sensor-generated time series constitute a large and growing portion of the world's data. Often, this data is collected from distributed, resource-constrained devices and centralized at one or more servers. A key challenge in this setup is reducing the size of the transmitted data without sacrificing its quality. Lower quality reduces the data's utility, but smaller size enables both reduced network and storage costs at the servers and reduced power consumption in sensing devices. A natural solution is to compress the data at the sensing devices. Unfortunately, existing compression algorithms either violate the memory and latency constraints common for these devices or, as we show experimentally, perform poorly on sensor-generated time series.We introduce a time series compression algorithm that achieves state-of-the-art compression ratios while requiring less than 1KB of memory and adding virtually no latency. This method is suitable not only for low-power devices collecting data, but also for servers storing and querying data; in the latter context, it can decompress at over 3GB/s in a single thread, even faster than many algorithms with much lower compression ratios. A key component of our method is a high-speed forecasting algorithm that can be trained online and significantly outperforms alternatives such as delta coding.Extensive experiments on datasets from many domains show that these results hold not only for sensor data but also across a wide array of other time series.

Intelligent organisms face a variety of tasks requiring the acquisition of expertise within a specific domain, including the ability to discriminate between a large number of similar patterns. From an energy-efficiency perspective, effective discrimination requires a prudent allocation of neural resources with more frequent patterns and their variants being represented with greater precision. In this work, we demonstrate a biologically plausible means of constructing a single-layer neural network that adaptively (i.e., without supervision) meets this criterion. Specifically, the adaptive algorithm includes synaptogenesis, synaptic shedding, and bi-directional synaptic weight modification to produce a network with outputs (i.e. neural codes) that represent input patterns proportional to the frequency of related patterns. In addition to pattern frequency, the correlational structure of the input environment also affects allocation of neural resources. The combined synaptic modification mechanisms provide an explanation of neuron allocation in the case of self-taught experts.

Robustness to certain forms of distribution shift is a key concern in many ML applications. Often, robustness can be formulated as enforcing invariances to particular interventions on the data generating process. Here, we study a flexible, causally-motivated approach to enforcing such invariances, paying special attention to shortcut learning, where a robust predictor can achieve optimal iid generalization in principle, but instead it relies on spurious correlations or shortcuts in practice. Our approach uses auxiliary labels, typically available at training time, to enforce conditional independences between the latent factors that determine these labels. We show both theoretically and empirically that causally-motivated regularization schemes (a) lead to more robust estimators that generalize well under distribution shift, and (b) have better finite sample efficiency compared to usual regularization schemes, even in the absence of distribution shifts. Our analysis highlights important theoretical properties of training techniques commonly used in causal inference, fairness, and disentanglement literature.