No abstract

Language models based on the Transformer architecture [1] have achieved state-ofthe-art performance on a wide range of natural language processing (NLP) tasks such as text classification, question-answering, and token classification. However, this performance is usually tested and reported on high-resource languages, like English, French, Spanish, and German. Indian languages, on the other hand, are underrepresented in such benchmarks. Despite some Indian languages being included in training multilingual Transformer models, they have not been the primary focus of such work. In order to evaluate the performance on Indian languages specifically, we analyze these language models through extensive experiments on multiple downstream tasks in Hindi, Bengali, and Telugu language. Here, we compare the efficacy of fine-tuning model parameters of pre-trained models against that of training a language model from scratch. Moreover, we empirically argue against the strict dependency between the dataset size and model performance, but rather encourage task-specific model and method selection. We achieve state-of-the-art performance on Hindi and Bengali languages for text classification task. Finally, we present effective strategies for handling the modeling of Indian languages and we release our model checkpoints for the community : https://huggingface.co/neuralspace-reverie. * Equal contribution Preprint. Under review.

No abstract

Using high-performance LC (E7) filled microfabricated refractive Fresnel chambers, we experimentally demonstrate a thin low-profile adaptive optical system with high analog tunability (2.1 D) that can be integrated with an adaptive contact-lens system.

No abstract

Smart, adaptive contact lenses (SCLs) are amongst the most anticipated, next-generation, standalone medical devices. SCLs require the integration of thin microelectronic components, tunable lenses, and micro-power sources onto a common non-planar substrate. Here, we report a miniaturized, sliding metalair electrochemical micro-battery driven by natural eye blinking motion that can be integrated with an SCL platform as a source of electrical energy. The metal-air battery (3⋅8 mm 2 ) consists of a Mg anode and a Pt cathode. The electrolyte of the battery is the eye-tear liquid and is introduced to the battery structure during the natural eye-blinking cycle, which activates the battery. The open-circuit voltage across the eyetear activated metal-air battery (ETMAB) was measured to be 2.2 V and the maximum speci c capacity of 3561 mA h g −1 was obtained at a discharge current density of 5 mA•cm −2 . Impedance matching analysis exhibits the maximum generated power density of 1.3 mW•cm −2 at the load of 740 Ω.

No abstract

Supplying electric power to wearable IoT devices, particularly smart contact lenses (SCLs), is one of the main obstacles to widespread adoption and commercialization. In the present study, we have successfully designed, fabricated, and characterized semi-transparent, self-supported, and flexible single crystalline silicon solar cells using a single-sided micromachining procedure. Optical, mechanical, and electrical simulations, together with the practical measurements, verify the application of our developed solar cells to be mounted on a limitedfootprint and flexible SCL. The 15 μm-thick silicon solar cells conformally fit on a dome-shaped contact lens (ROC = 8 mm) without any mechanical and electrical degradation. This homojunction photovoltaic device containing an array of microholes exhibits a V oc , J sc , and maximum power density of 504 mV, 6.48 mA cm −2 , and 1.67 mW cm −2 , respectively, at 25% visible light transparency under an AM1.5 one sun condition. Furthermore, the measurements were conducted under low-intensity indoor light conditions and resulted in a maximum power output of 25 and 42 μW cm −2 for the 50 and 25% transparent solar cells, respectively.