No abstract

Sharing knowledge for multiple related machine learning tasks is an effective strategy to improve the generalization performance. In this paper, we investigate knowledge sharing across categories for action recognition in videos. The motivation is that many action categories are related, where common motion pattern are shared among them (e.g. diving and high jump share the jump motion). We propose a new multi-task learning method to learn latent tasks shared across categories, and reconstruct a classifier for each category from these latent tasks. Compared to previous methods, our approach has two advantages: (1) The learned latent tasks correspond to basic motion patterns instead of full actions, thus enhancing discrimination power of the classifiers. (2) Categories are selected to share information with a sparsity regularizer, avoiding falsely forcing all categories to share knowledge. Experimental results on multiple public data sets show that the proposed approach can effectively transfer knowledge between different action categories to improve the performance of conventional single task learning methods.

Remotely monitoring changes in central U.S. grasslands is challenging because these landscapes tend to respond quickly to disturbances and changes in weather. Such dynamic responses influence nutrient cycling, greenhouse gas contributions, habitat availability for wildlife, and other ecosystem processes and services. Traditionally, coarse-resolution satellite data acquired at daily intervals have been used for monitoring. Recently, the harmonized Landsat-8 and Sentinel-2 (HLS) data increased the temporal frequency of the data. Here we investigated if the increased data frequency provided adequate observations to characterize highly dynamic grassland processes. We evaluated HLS data available for 2016 to (1) determine if data from Sentinel-2 contributed to an improvement in characterizing landscape processes over Landsat-8 data alone, and (2) quantify how observation frequency impacted results. Specifically, we investigated into estimating annual vegetation phenology, detecting burn scars from fire, and modeling within-season wetland hydroperiod and growth of aquatic vegetation. We observed increased sensitivity to the start of the growing season (SOST) with the HLS data. Our estimates of the grassland SOST compared well with ground estimates collected at a phenological camera site. We used the Continuous Change Detection and Classification (CCDC) algorithm to assess if the HLS data improved our detection of burn scars following grassland fires and found that detection was considerably influenced by the seasonal timing of the fires. The grassland burned in early spring recovered too quickly to be detected as change events by CCDC; instead, the spectral characteristics following these fires were incorporated as part of the ongoing time-series models. In contrast, the spectral effects from late-season fires were detected both by Landsat-8 data and HLS data. For wetland-rich areas, we used a modified version of the CCDC algorithm to track within-season dynamics of water and aquatic vegetation. The addition of Sentinel-2 data provided the potential to build full time series models to better distinguish different wetland types, suggesting that the temporal density of data was sufficient for within-season characterization of wetland dynamics. Although the different data frequency, in both the spatial and temporal dimensions, could cause inconsistent model estimation or sensitivity sometimes; overall, the temporal frequency of the HLS data improved our ability to track within-season grassland dynamics and improved results for areas prone to cloud contamination. The results suggest a greater frequency of observations, such as from harmonizing data across all comparable Landsat and Sentinel sensors, is still needed. For our study areas, at least a 3-day revisit interval during the early growing season (weeks 14–17) is required to provide a >50% probability of obtaining weekly clear observations.

Multi-task learning (MTL) methods have shown promising performance by learning multiple relevant tasks simultaneously, which exploits to share useful information across relevant tasks. Among various MTL methods, clustered multi-task learning (CMTL) assumes that all tasks can be clustered into groups and attempts to learn the underlying cluster structure from the training data. In this paper, we present a new approach for CMTL, called flexible clustered multi-task (FCMTL), in which the cluster structure is learned by identifying representative tasks. The new approach allows an arbitrary task to be described by multiple representative tasks, effectively soft-assigning a task to multiple clusters with different weights. Unlike existing counterpart, the proposed approach is more flexible in that (a) it does not require clusters to be disjoint, (b) tasks within one particular cluster do not have to share information to the same extent, and (c) the number of clusters is automatically inferred from data. Computationally, the proposed approach is formulated as a row-sparsity pursuit problem. We validate the proposed FCMTL on both synthetic and real-world data sets, and empirical results demonstrate that it outperforms many existing MTL methods.