RESUMEN
Mixed-precision quantization mostly predetermines the model bit-width settings before actual training due to the non-differential bit-width sampling process, obtaining suboptimal performance. Worse still, the conventional static quality-consistent training setting, i.e., all data is assumed to be of the same quality across training and inference, overlooks data quality changes in real-world applications which may lead to poor robustness of the quantized models. In this article, we propose a novel data quality-aware mixed-precision quantization framework, dubbed DQMQ, to dynamically adapt quantization bit-widths to different data qualities. The adaption is based on a bit-width decision policy that can be learned jointly with the quantization training. Concretely, DQMQ is modeled as a hybrid reinforcement learning (RL) task that combines model-based policy optimization with supervised quantization training. By relaxing the discrete bit-width sampling to a continuous probability distribution that is encoded with few learnable parameters, DQMQ is differentiable and can be directly optimized end-to-end with a hybrid optimization target considering both task performance and quantization benefits. Trained on mixed-quality image datasets, DQMQ can implicitly select the most proper bit-width for each layer when facing uneven input qualities. Extensive experiments on various benchmark datasets and networks demonstrate the superiority of DQMQ against existing fixed/mixed-precision quantization methods.
RESUMEN
Narrow bandwidths are a general bottleneck for applications relying on passive, linear, subwavelength resonators. In the past decades, several efforts have been devoted to overcoming this challenge, broadening the bandwidth of small resonators by the means of analog non-Foster matching networks for radiators, antennas and metamaterials. However, most non-Foster approaches present challenges in terms of tunability, stability and power limitations. Here, by tuning a subwavelength acoustic transducer with digital non-Foster-inspired electronics, we demonstrate five-fold bandwidth enhancement compared to conventional analog non-Foster matching. Long-distance transmission over airborne acoustic channels, with approximately three orders of magnitude increase in power level, validates the performance of the proposed approach. We also demonstrate convenient reconfigurability of our non-Foster-inspired electronics. This implementation provides a viable solution to enhance the bandwidth of sub-wavelength resonance-based systems, extendable to the electromagnetic domain, and enables the practical implementation of airborne and underwater acoustic radiators.
RESUMEN
Succinate dehydrogenase inhibitors are essential fungicides used in agriculture. To explore new pyrazole-carboxamides with high fungicidal activity, a series of N-substitutedphenyl-3-di/trifluoromethyl-1-methyl-1H-pyrazole-4-carboxamides bearing a branched alkyl ether moiety were designed and synthesized. The in vitro bioassay indicated that some target compounds displayed appreciable fungicidal activity. For example, compounds 5d and 5e showed high efficacy against S. sclerotiorum with EC50 values of 3.26 and 1.52 µg/mL respectively, and also exhibited excellent efficacy against R. solani with EC50 values of 0.27 and 0.06 µg/mL respectively, which were comparable or superior to penflufen. The further in vivo bioassay on cucumber leaves demonstrated that 5e provided strong protective activity of 94.3 % against S. sclerotiorum at 100 µg/mL, comparable to penflufen (99.1 %). Cytotoxicity assessment against human renal cell lines (239A cell) revealed that 5e had low cytotoxicity within the median effective concentrations. Docking study of 5e with succinate dehydrogenase illustrated that R-5e formed one hydrogen bond and two π-π stacking interactions with amino acid residues of target enzyme, while S-5e formed only one π-π stacking interaction with amino acid residue. This study provides a valuable reference for the design of new succinate dehydrogenase inhibitor.