In this paper we present for the first time a framework that allows secure two-party computations on approximations of real valued signals. In our solution, we use a quantized logarithmic representation of the signal samples, which enables to represent both very small and very large numbers with bounded relative error. We show that numbers represented in this way can be encrypted using standard homomorphic encryption schemes; furthermore we give protocols that allow to perform all arithmetic operations on such encrypted values. Finally we demonstrate the practicality of our framework by applying it to the problem of filtering encrypted signals.

No abstract

No abstract

Massively multilingual sentence representations are trained on large corpora of uncurated data, with a very imbalanced proportion of languages included in the training. This may cause the models to grasp cultural values including moral judgments from the high-resource languages and impose them on the low-resource languages. The lack of data in certain languages can also lead to developing random and thus potentially harmful beliefs. Both these issues can negatively influence zero-shot cross-lingual model transfer and potentially lead to harmful outcomes. Therefore, we aim to (1) detect and quantify these issues by comparing different models in different languages, (2) develop methods for improving undesirable properties of the models. Our initial experiments using the multilingual model XLM-R show that indeed multilingual LMs capture moral norms, even with potentially higher human-agreement than monolingual ones. However, it is not yet clear to what extent these moral norms differ between languages.Recent work demonstrated large pre-trained language models (PLM) obtain symbolic, relational [12] but also commonsense knowledge [5]. Further, West et al. [17] showed that one is able to extract the commonsense knowledge from the large, general language model GPT-3 [2] via symbolic knowledge distillation. This encoded "knowledge" includes information of our society reflecting ethical

Text-conditioned image generation models have recently achieved astonishing results in image quality and text alignment and are consequently employed in a fast-growing number of applications. Since they are highly data-driven, relying on billion-sized datasets randomly scraped from the internet, they also suffer, as we demonstrate, from degenerated and biased human behavior. In turn, they may even reinforce such biases. To help combat these undesired side effects, we present safe latent diffusion (SLD). Specifically, to measure the inappropriate degeneration due to unfiltered and imbalanced training sets, we establish a novel image generation test bed-inappropriate image prompts (I2P)-containing dedicated, real-world image-totext prompts covering concepts such as nudity and violence. As our exhaustive empirical evaluation demonstrates, the introduced SLD removes and suppresses inappropriate image parts during the diffusion process, with no additional training required and no adverse effect on overall image quality or text alignment.