Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study introduces Barnamala, a compact convolutional network with only 1.11 million parameters, achieving a remarkable 99.73% accuracy in Devanagari handwritten character recognition, which is the highest reported performance while being 15.6 times smaller than the previous state-of-the-art. The authors identify a saturation point in model performance, where even large teacher ensembles fail to surpass an intrinsic error floor of 11 errors, as confirmed by rigorous statistical testing. Furthermore, Barnamala demonstrates competitive performance in zero-shot and fine-tuning scenarios on external datasets, showcasing superior robustness against data corruption compared to larger models.
Achieving 99.73% accuracy in Devanagari recognition with a model 15.6x smaller than the previous best highlights a critical saturation point in performance that challenges conventional scaling assumptions.
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.