发表机构
Ampixa Labs; Pulchowk Campus（安皮克萨实验室; 普尔乔克校区）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究针对手写天城体识别，构建紧凑型卷积网络，在46类DHCD识别中达99.73%准确率，比此前先进模型小15.6倍。模型达饱和点，学生模型即便无知识蒸馏也能匹配大型基线，在其他数据集上零样本和微调表现佳，抗干扰能力强。
AI 中文摘要
我们构建了一个用于46类DHCD天城体识别的紧凑型卷积网络（111万个参数），准确率达到99.73%，比之前的最先进水平小15.6倍且是已报道的最高值。我们有效达到了饱和点，所有测试模型（包括大型教师模型集成）都达到相同的11错误内在下限。在精确的McNemar检验和威尔逊置信区间下，没有配置能在统计上显著胜出。即使没有知识蒸馏，我们的学生模型也能与最近的大型模型基线匹配。在DHCD之外，对CMATERdb数字进行零样本学习准确率为76.6%，微调后达到97.8%；抗干扰能力也远优于大型基线（平均抗干扰准确率75.7%对38.7%）。所有工件可在该https网址获取。
英文摘要
We built a compact convolutional network (1.11 M parameters) for 46-class DHCD Devanagari recognition and reached 99.73%, the highest reported at 15.6x smaller than prior state-of-the-art. We have effectively reached the saturation point: every model tested, large teacher ensembles included, hits the same 11-error intrinsic floor. No configuration achieves a statistically clear win under exact McNemar tests with Wilson confidence intervals. Even without knowledge distillation, our student matches the nearest large-model baseline (17.32 M parameters; McNemar $p = 0.345$). Outside of DHCD, zero-shot on CMATERdb digits gives 76.6% and fine-tuning reaches 97.8%; corruption robustness is also far better than large baselines (mean corruption accuracy 75.7% vs. 38.7%). All artifacts are at https://github.com/Ampixa/barnamala.
Comments14 pages, 2 figures, 7 tables. Code and artifacts available at https://github.com/Ampixa/barnamala
