Homomorphic encryption (HE) is the ultimate tool for performing secure computations even in untrusted environments. Application of HE for deep learning (DL) inference is an active area of research, given the fact that DL models are often deployed in untrusted environments (e.g., third-party servers) yet inferring on private data. However, existing HE libraries [somewhat (SWHE), leveled (LHE) or fully homomorphic (FHE)] suffer from extensive computational and memory overhead. Few performance optimized high-speed homomorphic libraries are either suffering from certain approximation issues leading to decryption errors or proven to be insecure according to recent published attacks. In this article, we propose architectural tricks to achieve performance speedup for encrypted DL inference developed with exact HE schemes without any approximation or decryption error in homomorphic computations. The main idea is to apply quantization and suitable data packing in the form of bitslicing to reduce the costly noise handling operation, Bootstrapping while achieving a functionally correct and highly parallel DL pipeline with a moderate memory footprint. Experimental evaluation on the MNIST dataset shows a significant ( $37\times$ ) speedup over the nonbitsliced versions of the same architecture. Low memory bandwidths (700 MB) of our design pipelines further highlight their promise toward scaling over larger gamut of Edge-AI analytics use cases.

Evaluation of the Special Sensor Microwave Imager (SSM/I) precipitation features is presented from various passes over Indian oceanic and land regions under the framework of radiative transfer simulations for both emission and scattering atmospheres. There is considerable uncertainty in the interpretation of the SSM/I high-frequency scattering features for development of rainfall algorithms. Specifically large areas of very low emissivity regimes showing false rain signatures due to the presence of low brightness temperatures (TBs) that are often present in the vicinity of colder sea surface areas in the 85-GHz TB from SSM/I. In this connection, the Polarization Corrected Temperature (PCT) defined by Spencer using the 85-GHz channels, vertical (V) and horizontal (H), has been studied to delineate these surface effects. These false scattering signatures, once corrected using PCT as a suitable parameter, significantly improve the quality of the SSM/I-derived precipitation areas. These results are also confirmed with the cloud optical depth data from the Moderate Resolution Imaging Spectroradiometer sensor onboard the Aqua satellite and the standard merged rain product (geostationary infrared and Tropical Rainfall Measuring Mission (TRMM) microwave data) 3B42 from TRMM. This letter is aimed for selecting PCT as one of the suitable predictor variables for the development of operational rainfall retrieval algorithm for the Indo-French Megha-Tropiques satellite's microwave radiometer at high frequencies.

In 2004, the Tropical Rainfall Measuring Mission (TRMM) Science Project released a newer version of precipitation products, i.e., version 6 (V6), from its various instruments. The V6 data products are expected to be more accurate than the previous version 5 (V5). In this letter, we have attempted to analyze V5 and V6 products from two primary sensors on the TRMM, namely, the TRMM Microwave Imager (TMI) and the Precipitation Radar (PR), to unravel the quality of the V6 products vis-a-vis that of the V5 products. It is found that there are significant changes in the TMI-derived precipitation values, but the TMI brightness temperature (Tb) has not undergone any change from V5 to V6. Thus, the retrieval algorithm must have undergone some changes from V5 to V6. While the total number of TMI raining pixels is nearly unchanged, V5 moderate rain events (1 - 5 mm ldr h-1) are more often classified as low rain events (-1) in V6. Thus, TMI-based precipitation shows larger bias, particularly at low to moderate rain rates than that in V5. This results in a depression in the average Tb at around 2 mm ldr h-1in the Tb-rain-rate relationship with V6 measurements, which is implausible. This dip in average Tb value further complicates the Tb-rain-rate relationship by showing a continuous rise in Tb with rain rate, which is again implausible. On the other hand, PR-based rain rates have not undergone much change from V5 to V6. The relationship between Tb from TMI and rain rate from PR also does not show any anomalous behavior.