Estimating the out-of-distribution performance in regimes where labels are scarce is critical to safely deploy foundation models. Recently, it was shown that ensembles of neural networks observe the phenomena "agreement-on-the-line", which can be leveraged to reliably predict OOD performance without labels. However, in contrast to classical neural networks that are trained on in-distribution data from scratch for numerous epochs, foundation models undergo minimal finetuning from heavily pretrained weights, which may reduce the ensemble diversity needed to observe agreement-on-the-line. In our work, we demonstrate that when lightly finetuning multiple runs from a $\textit{single}$ foundation model, the choice of randomness during training (linear head initialization, data ordering, and data subsetting) can lead to drastically different levels of agreement-on-the-line in the resulting ensemble. Surprisingly, only random head initialization is able to reliably induce agreement-on-the-line in finetuned foundation models across vision and language benchmarks. Second, we demonstrate that ensembles of $\textit{multiple}$ foundation models pretrained on different datasets but finetuned on the same task can also show agreement-on-the-line. In total, by careful construction of a diverse ensemble, we can utilize agreement-on-the-line-based methods to predict the OOD performance of foundation models with high precision.

Due to increased data movement and information exchange over internet and web, preserving data confidentiality and security has emerged as a prime concern for the end users. From bank transactions to document verification portals, from government official websites to social media, all these platforms share data remotely over web which contains certain confidential information and thus uses security mechanisms based on cryptographic algorithm to preserve the confidentiality of data preventing any information leakage or breach. RSA being the most popular public key cryptography based algorithm uses data encryption and decryption technique. It uses the mathematical concept of factorization of prime integers. The efficiency of the algorithm in terms of enhancing the security aspect to guess the factors of prime numbers increase when the numbers have high powers. Computing prime factors of these large numbers is compute intensive task where the serial programming makes RSA algorithm to slow down. The paper here presents an OpenMP based algorithmic modification to the code of RSA algorithm to improve the execution time using the parallel processing power of modern day multi-core architecture based machines. Experimental results under the lights of graphical representation shows a considerable speed up gained over the traditional implementation.