Not all video frames are equally informative for recognizing an action. It is
computationally infeasible to train deep networks on all video frames when
actions develop over hundreds of frames. A common heuristic is uniformly
sampling a small number of video frames and using these to recognize the
action. Instead, here we propose full video action recognition and consider all
video frames. To make this computational tractable, we first cluster all frame
activations along the temporal dimension based on their similarity with respect
to the classification task, and then temporally aggregate the frames in the
clusters into a smaller number of representations. Our method is end-to-end
trainable and computationally efficient as it relies on temporally localized
clustering in combination with fast Hamming distances in feature space. We
evaluate on UCF101, HMDB51, Breakfast, and Something-Something V1 and V2, where
we compare favorably to existing heuristic frame sampling methods.

通过对视频帧进行时间聚类，提出了一种全视频行为识别方法，相较于现有的基于帧采样的方法效果更好，同时由于采用了时空局部聚类和快速的哈明距离计算方法使其训练高效。

无遗漏：全视频动作识别

No frame left behind: Full Video Action Recognition

Human behavior drives a range of complex social, urban, and economic systems,
yet understanding its structure and dynamics at the individual level remains an
open question. From credit card transactions to communications data, human
behavior appears to exhibit bursts of activity driven by task prioritization
and periodicity, however, current research does not offer generative models
capturing these mechanisms. We propose a multivariate, periodic Hawkes process
(MPHP) model that captures -- at the individual level -- the temporal
clustering of human activity, the interdependence structure and co-excitation
of different activities, and the periodic effects of weekly rhythms. We also
propose a scalable parameter estimation technique for this model using
maximum-aposteriori expectation-maximization that additionally provides
estimation of latent variables revealing branching structure of an individual's
behavior patterns. We apply the model to a large dataset of credit card
transactions, and demonstrate the MPHP outperforms a non-homogeneous Poisson
model and LDA in both statistical fit for the distribution of inter-event times
and an activity prediction task.

通过提出一种作为生成模型的多变量周期 Hawkes 过程模型，该模型可以捕捉人类活动的时间聚类、相互依赖结构和相互兴奋以及一周内时间周期效应。我们将这个模型应用于大规模信用卡交易数据集，并证明 MPHP 在交互事件时间分布和活动预测任务的统计适合度上超过非齐次泊松模型和 LDA。

利用数字足迹对人类动态和生活方式建模

Modeling Human Dynamics and Lifestyle Using Digital Traces

Electronic records contain sequences of events, some of which take place all
at once in a single visit, and others that are dispersed over multiple visits,
each with a different timestamp. We postulate that fine temporal detail, e.g.,
whether a series of blood tests are completed at once or in rapid succession
should not alter predictions based on this data. Motivated by this intuition,
we propose models for analyzing sequences of multivariate clinical time series
data that are invariant to this temporal clustering. We propose an efficient
data augmentation technique that exploits the postulated temporal-clustering
invariance to regularize deep neural networks optimized for several clinical
prediction tasks. We introduce two techniques to temporally coarsen
(downsample) irregular time series: (i) grouping the data points based on
regularly-spaced timestamps; and (ii) clustering them, yielding
irregularly-paced timestamps. Moreover, we propose a MultiResolution Ensemble
(MRE) model, improving predictive accuracy by ensembling predictions based on
inputs sequences transformed by different coarsening operators. Our experiments
show that MRE improves the mAP on the benchmark mortality prediction task from
51.53% to 53.92%.

本研究基于数据时间聚类的不变性，提出了分析多变量临床时间序列数据的模型以及一种数据扩充技术方案，用于规范化深度神经网络中医学预测任务，该方法以评估医疗预测任务为基准，提高了其预测精度。