Search for: All records

While existing hierarchical text classification (HTC) methods attempt to capture label hierarchies for model training, they either make local decisions regarding each label or completely ignore the hierarchy information during inference. To solve the mismatch between training and inference as well as modeling label dependencies in a more principled way, we formulate HTC as a Markov decision process and propose to learn a Label Assignment Policy via deep reinforcement learning to determine where to place an object and when to stop the assignment process. The proposed method, HiLAP, explores the hierarchy during both training and inference time in a consistent manner and makes inter-dependent decisions. As a general framework, HiLAP can incorporate different neural encoders as base models for end-to-end training. Experiments on five public datasets and four base models show that HiLAP yields an average improvement of 33.4% in Macro-F1 over flat classifiers and outperforms state-of-the-art HTC methods by a large margin. 

The profiles of marine boundary layer (MBL) cloud and drizzle microphysical properties are important for studying the cloud‐to‐rain conversion and growth processes in MBL clouds. However, it is challenging to simultaneously retrieve both cloud and drizzle microphysical properties within an MBL cloud layer using ground‐based observations. In this study, methods were developed to first decompose drizzle and cloud reflectivity in MBL clouds from Atmospheric Radiation Measurement cloud radar reflectivity measurements and then simultaneously retrieve cloud and drizzle microphysical properties during the Aerosol and Cloud Experiments in the Eastern North Atlantic (ACE‐ENA) campaign. These retrieved microphysical properties, such as cloud and drizzle particle size (r_candr_m,d), their number concentration (N_candN_d) and liquid water content (LWC_candLWC_d), have been validated by aircraft in situ measurements during ACE‐ENA (~158 hr of aircraft data). The mean surface retrieved (in situ measured)r_c,N_c, andLWC_care 10.9 μm (11.8 μm), 70 cm⁻³(60 cm⁻³), and 0.21 g m⁻³(0.22 g m⁻³), respectively. For drizzle microphysical properties, the retrieved (in situ measured)r_d,N_d, andLWC_dare 44.9 μm (45.1 μm), 0.07 cm⁻³(0.08 cm⁻³), and 0.052 g m⁻³(0.066 g m⁻³), respectively. Treating the aircraft in situ measurements as truth, the estimated median retrieval errors are ~15% forr_c, ~35% forN_c, ~30% forLWC_candr_d, and ~50% forN_dandLWC_d. The findings from this study will provide insightful information for improving our understanding of warm rain processes, as well as for improving model simulations. More studies are required over other climatic regions.