Search

The effective detection of evidence of financial anomalies requires collaboration among multiple entities who own a diverse set of data, such as a payment network system (PNS) and its partner banks. Trust among these financial institutions is limited by regulation and competition. Federated learning (FL) enables entities to collaboratively train a model when data is either vertically or horizontally partitioned across the entities. However, in real-world financial anomaly detection scenarios, the data is partitioned both vertically and horizontally and hence it is not possible to use existing FL approaches in a plug-and-play manner. Our novel solution, PV4FAD, combines fully homomorphic encryption (HE), secure multi-party computation (SMPC), differential privacy (DP), and randomization techniques to balance privacy and accuracy during training and to prevent inference threats at model deployment time. Our solution provides input privacy through HE and SMPC, and output privacy against inference time attacks through DP. Specifically, we show that, in the honest-but-curious threat model, banks do not learn any sensitive features about PNS transactions, and the PNS does not learn any information about the banks' dataset but only learns prediction labels. We also develop and analyze a DP mechanism to protect output privacy during inference. Our solution generates high-utility models by significantly reducing the per-bank noise level while satisfying distributed DP. To ensure high accuracy, our approach produces an ensemble model, in particular, a random forest. This enables us to take advantage of the well-known properties of ensembles to reduce variance and increase accuracy. Our solution won second prize in the first phase of the U.S. Privacy Enhancing Technologies (PETs) Prize Challenge., Comment: Prize Winner in the U.S. Privacy Enhancing Technologies (PETs) Prize Challenge

Text-to-image diffusion models achieved a remarkable leap in capabilities over the last few years, enabling high-quality and diverse synthesis of images from a textual prompt. However, even the most advanced models often struggle to precisely follow all of the directions in their prompts. The vast majority of these models are trained on datasets consisting of (image, caption) pairs where the images often come from the web, and the captions are their HTML alternate text. A notable example is the LAION dataset, used by Stable Diffusion and other models. In this work we observe that these captions are often of low quality, and argue that this significantly affects the model's capability to understand nuanced semantics in the textual prompts. We show that by relabeling the corpus with a specialized automatic captioning model and training a text-to-image model on the recaptioned dataset, the model benefits substantially across the board. First, in overall image quality: e.g. FID 14.84 vs. the baseline of 17.87, and 64.3% improvement in faithful image generation according to human evaluation. Second, in semantic alignment, e.g. semantic object accuracy 84.34 vs. 78.90, counting alignment errors 1.32 vs. 1.44 and positional alignment 62.42 vs. 57.60. We analyze various ways to relabel the corpus and provide evidence that this technique, which we call RECAP, both reduces the train-inference discrepancy and provides the model with more information per example, increasing sample efficiency and allowing the model to better understand the relations between captions and images.

Text-to-SQL semantic parsing faces challenges in generalizing to cross-domain and complex queries. Recent research has employed a question decomposition strategy to enhance the parsing of complex SQL queries. However, this strategy encounters two major obstacles: (1) existing datasets lack question decomposition; (2) due to the syntactic complexity of SQL, most complex queries cannot be disentangled into sub-queries that can be readily recomposed. To address these challenges, we propose a new modular Query Plan Language (QPL) that systematically decomposes SQL queries into simple and regular sub-queries. We develop a translator from SQL to QPL by leveraging analysis of SQL server query optimization plans, and we augment the Spider dataset with QPL programs. Experimental results demonstrate that the modular nature of QPL benefits existing semantic-parsing architectures, and training text-to-QPL parsers is more effective than text-to-SQL parsing for semantically equivalent queries. The QPL approach offers two additional advantages: (1) QPL programs can be paraphrased as simple questions, which allows us to create a dataset of (complex question, decomposed questions). Training on this dataset, we obtain a Question Decomposer for data retrieval that is sensitive to database schemas. (2) QPL is more accessible to non-experts for complex queries, leading to more interpretable output from the semantic parser., Comment: EMNLP 2023 Findings

Frequency mixing processes in spin systems have a variety of applications in meteorology and in quantum data processing. Spin spectroscopy based on frequency mixing offers some advantages, including the ability to eliminate crosstalk between driving and detection. We experimentally explore nonlinear frequency mixing processes with negatively charged nitrogen-vacancy defects in diamond at low temperatures, and near level anti crossing. The experimental setup allows simultaneously applying magnetic driving in the longitudinal and transverse directions. Magnetic resonance detection is demonstrated using both Landau Zener St\"uckelberg interferometry and two-tone driving spectroscopy. The experimental results are compared with predictions of a theoretical analysis based on the rotating wave approximation.

Every day, humans perceive objects and communicate these perceptions through various channels. In this paper, we present a computational model designed to track and simulate the perception of objects, as well as their representations as conveyed in communication. We delineate two fundamental components of our internal representation, termed "observed" and "seen", which we correlate with established concepts in computer vision, namely encoding and decoding. These components are integrated into semiotic networks, which simulate perceptual inference of object perception and human communication. Our model of object perception by a person allows us to define object perception by {\em a network}. We demonstrate this with an example of an image baseline classifier by constructing a new network that includes the baseline classifier and an additional layer. This layer produces the images "perceived" by the entire network, transforming it into a perceptualized image classifier. This facilitates visualization of the acquired network. Within our network, the image representations become more efficient for classification tasks when they are assembled and randomized. In our experiments, the perceptualized network outperformed the baseline classifier on MNIST training databases consisting of a restricted number of images. Our model is not limited to persons and can be applied to any system featuring a loop involving the processing from "internal" to "external" representations.

Text-To-Image (TTI) models, such as DALL-E and StableDiffusion, have demonstrated remarkable prompt-based image generation capabilities. Multilingual encoders may have a substantial impact on the cultural agency of these models, as language is a conduit of culture. In this study, we explore the cultural perception embedded in TTI models by characterizing culture across three hierarchical tiers: cultural dimensions, cultural domains, and cultural concepts. Based on this ontology, we derive prompt templates to unlock the cultural knowledge in TTI models, and propose a comprehensive suite of evaluation techniques, including intrinsic evaluations using the CLIP space, extrinsic evaluations with a Visual-Question-Answer (VQA) model and human assessments, to evaluate the cultural content of TTI-generated images. To bolster our research, we introduce the CulText2I dataset, derived from six diverse TTI models and spanning ten languages. Our experiments provide insights regarding Do, What, Which and How research questions about the nature of cultural encoding in TTI models, paving the way for cross-cultural applications of these models., Comment: Project page: https://venturamor.github.io/CulText2IWeb/

We study the low temperature $(2+1)$D Solid-On-Solid model on $[[1, L ]]^2$ with zero boundary conditions and nonnegative heights (a floor at height $0$). Caputo et al. (2016) established that this random surface typically admits either $\mathfrak h $ or $\mathfrak h+1$ many nested macroscopic level line loops $\{\mathcal L_i\}_{i\geq 0}$ for an explicit $\mathfrak h\asymp \log L$, and its top loop $\mathcal L_0$ has cube-root fluctuations: e.g., if $\rho(x)$ is the vertical displacement of $\mathcal L_0$ from the bottom boundary point $(x,0)$, then $\max \rho(x) = L^{1/3+o(1)}$ over $x\in I_0:=L/2+[[-L^{2/3},L^{2/3}]]$. It is believed that rescaling $\rho$ by $L^{1/3}$ and $I_0$ by $L^{2/3}$ would yield a limit law of a diffusion on $[-1,1]$. However, no nontrivial lower bound was known on $\rho(x)$ for a fixed $x\in I_0$ (e.g., $x=\frac L2$), let alone on $\min\rho(x)$ in $I_0$, to complement the bound on $\max\rho(x)$. Here we show a lower bound of the predicted order $L^{1/3}$: for every $\epsilon>0$ there exists $\delta>0$ such that $\min_{x\in I_0} \rho(x) \geq \delta L^{1/3}$ with probability at least $1-\epsilon$. The proof relies on the Ornstein--Zernike machinery due to Campanino-Ioffe-Velenik, and a result of Ioffe, Shlosman and Toninelli (2015) that rules out pinning in Ising polymers with modified interactions along the boundary. En route, we refine the latter result into a Brownian excursion limit law, which may be of independent interest. We further show that in a $ K L^{2/3}\times K L^{2/3}$ box with boundary conditions $\mathfrak h-1,\mathfrak h,\mathfrak h,\mathfrak h$ (i.e., $\mathfrak h-1$ on the bottom side and $\mathfrak h$ elsewhere), the limit of $\rho(x)$ as $K,L\to\infty$ is a Ferrari--Spohn diffusion., Comment: 63 pages, 2 figures. Final version

We introduce a distinctive real-time, causal, neural network-based active speaker detection system optimized for low-power edge computing. This system drives a virtual cinematography module and is deployed on a commercial device. The system uses data originating from a microphone array and a 360-degree camera. Our network requires only 127 MFLOPs per participant, for a meeting with 14 participants. Unlike previous work, we examine the error rate of our network when the computational budget is exhausted, and find that it exhibits graceful degradation, allowing the system to operate reasonably well even in this case. Departing from conventional DOA estimation approaches, our network learns to query the available acoustic data, considering the detected head locations. We train and evaluate our algorithm on a realistic meetings dataset featuring up to 14 participants in the same meeting, overlapped speech, and other challenging scenarios.

An adjacency-crossing graph is a graph that can be drawn such that every two edges that cross the same edge share a common endpoint. We show that the number of edges in an $n$-vertex adjacency-crossing graph is at most $5n-10$. If we require the edges to be drawn as straight-line segments, then this upper bound becomes $5n-11$. Both of these bounds are tight. The former result also follows from a very recent and independent work of Cheong et al.\cite{cheong2023weakly} who showed that the maximum size of weakly and strongly fan-planar graphs coincide. By combining this result with the bound of Kaufmann and Ueckerdt\cite{KU22} on the size of strongly fan-planar graphs and results of Brandenburg\cite{Br20} by which the maximum size of adjacency-crossing graphs equals the maximum size of fan-crossing graphs which in turn equals the maximum size of weakly fan-planar graphs, one obtains the same bound on the size of adjacency-crossing graphs. However, the proof presented here is different, simpler and direct., Comment: 17 pages, 11 figures

It has been established that training a box-based detector network can enhance the localization performance of weakly supervised and unsupervised methods. Moreover, we extend this understanding by demonstrating that these detectors can be utilized to improve the original network, paving the way for further advancements. To accomplish this, we train the detectors on top of the network output instead of the image data and apply suitable loss backpropagation. Our findings reveal a significant improvement in phrase grounding for the ``what is where by looking'' task, as well as various methods of unsupervised object discovery. Our code is available at https://github.com/eyalgomel/box-based-refinement.

We consider an offline learning problem for an agent who first estimates an unknown price impact kernel from a static dataset, and then designs strategies to liquidate a risky asset while creating transient price impact. We propose a novel approach for a nonparametric estimation of the propagator from a dataset containing correlated price trajectories, trading signals and metaorders. We quantify the accuracy of the estimated propagator using a metric which depends explicitly on the dataset. We show that a trader who tries to minimise her execution costs by using a greedy strategy purely based on the estimated propagator will encounter suboptimality due to so-called spurious correlation between the trading strategy and the estimator and due to intrinsic uncertainty resulting from a biased cost functional. By adopting an offline reinforcement learning approach, we introduce a pessimistic loss functional taking the uncertainty of the estimated propagator into account, with an optimiser which eliminates the spurious correlation, and derive an asymptotically optimal bound on the execution costs even without precise information on the true propagator. Numerical experiments are included to demonstrate the effectiveness of the proposed propagator estimator and the pessimistic trading strategy., Comment: 12 figures

We introduce a function of the density of states for periodic Jacobi matrices on trees and prove a useful formula for it. This allows new, streamlined proofs of the gap labeling and Aomoto index theorems. We prove a version of this new formula for the Anderson model on trees.

We present Colordag, a blockchain protocol where following the prescribed strategy is, with high probability, a best response as long as all miners have less than 1/2 of the mining power. We prove the correctness of Colordag even if there is an extremely powerful adversary who knows future actions of the scheduler: specifically, when agents will generate blocks and when messages will arrive. The state-of-the-art protocol, Fruitchain, is an epsilon-Nash equilibrium as long as all miners have less than 1/2 of the mining power. However, there is a simple deviation that guarantees that deviators are never worse off than they would be by following Fruitchain, and can sometimes do better. Thus, agents are motivated to deviate. Colordag implements a solution concept that we call epsilon-sure Nash equilibrium and does not suffer from this problem. Because it is an epsilon-sure Nash equilibrium, Colordag is an epsilon Nash equilibrium and with probability (1 - epsilon) is a best response., Comment: To be published in DISC 2023

T1 mapping is a quantitative magnetic resonance imaging (qMRI) technique that has emerged as a valuable tool in the diagnosis of diffuse myocardial diseases. However, prevailing approaches have relied heavily on breath-hold sequences to eliminate respiratory motion artifacts. This limitation hinders accessibility and effectiveness for patients who cannot tolerate breath-holding. Image registration can be used to enable free-breathing T1 mapping. Yet, inherent intensity differences between the different time points make the registration task challenging. We introduce PCMC-T1, a physically-constrained deep-learning model for motion correction in free-breathing T1 mapping. We incorporate the signal decay model into the network architecture to encourage physically-plausible deformations along the longitudinal relaxation axis. We compared PCMC-T1 to baseline deep-learning-based image registration approaches using a 5-fold experimental setup on a publicly available dataset of 210 patients. PCMC-T1 demonstrated superior model fitting quality (R2: 0.955) and achieved the highest clinical impact (clinical score: 3.93) compared to baseline methods (0.941, 0.946 and 3.34, 3.62 respectively). Anatomical alignment results were comparable (Dice score: 0.9835 vs. 0.984, 0.988). Our code and trained models are available at https://github.com/eyalhana/PCMC-T1., Comment: Accepted to MICCAI 2023

The increasing versatility of language models (LMs) has given rise to a new class of benchmarks that comprehensively assess a broad range of capabilities. Such benchmarks are associated with massive computational costs, extending to thousands of GPU hours per model. However, the efficiency aspect of these evaluation efforts had raised little discussion in the literature. In this work, we present the problem of Efficient Benchmarking, namely, intelligently reducing the computation costs of LM evaluation without compromising reliability. Using the HELM benchmark as a test case, we investigate how different benchmark design choices affect the computation-reliability trade-off. We propose to evaluate the reliability of such decisions, by using a new measure -- Decision Impact on Reliability, DIoR for short. We find, for example, that a benchmark leader may change by merely removing a low-ranked model from the benchmark, and observe that a correct benchmark ranking can be obtained by considering only a fraction of the evaluation examples. Based on our findings, we outline a set of concrete recommendations for efficient benchmark design and utilization practices. To take a step further, we use our findings to propose an evaluation algorithm, that, when applied to the HELM benchmark, leads to dramatic cost savings with minimal loss of benchmark reliability, often reducing computation by x100 or more., Comment: Accepted to NAACL main track

The shortest path problem in graphs is fundamental to AI. Nearly all variants of the problem and relevant algorithms that solve them ignore edge-weight computation time and its common relation to weight uncertainty. This implies that taking these factors into consideration can potentially lead to a performance boost in relevant applications. Recently, a generalized framework for weighted directed graphs was suggested, where edge-weight can be computed (estimated) multiple times, at increasing accuracy and run-time expense. We build on this framework to introduce the problem of finding the tightest admissible shortest path (TASP); a path with the tightest suboptimality bound on the optimal cost. This is a generalization of the shortest path problem to bounded uncertainty, where edge-weight uncertainty can be traded for computational cost. We present a complete algorithm for solving TASP, with guarantees on solution quality. Empirical evaluation supports the effectiveness of this approach., Comment: arXiv admin note: text overlap with arXiv:2208.11489

Nakamoto's consensus protocol works in a permissionless model and tolerates Byzantine failures, but only offers probabilistic agreement. Recently, the Sandglass protocol has shown such weaker guarantees are not a necessary consequence of a permissionless model; yet, Sandglass only tolerates benign failures, and operates in an unconventional partially synchronous model. We present Gorilla Sandglass, the first Byzantine tolerant consensus protocol to guarantee, in the same synchronous model adopted by Nakamoto, deterministic agreement and termination with probability 1 in a permissionless setting. We prove the correctness of Gorilla by mapping executions that would violate agreement or termination in Gorilla to executions in Sandglass, where we know such violations are impossible. Establishing termination proves particularly interesting, as the mapping requires reasoning about infinite executions and their probabilities., Comment: 43 pages, 3 figures, published in the International Symposium on Distributed Computing (DISC) 2023

We study Glauber dynamics for the low temperature $(2+1)$D Solid-On-Solid model on a box of side-length $n$ with a floor at height $0$ (inducing entropic repulsion) and a competing bulk external field $\lambda$ pointing down (the prewetting problem). In 1996, Cesi and Martinelli showed that if the inverse-temperature $\beta$ is large enough, then along a decreasing sequence of critical points $(\lambda_c^{(k)})_{k=0}^{K_\beta}$ the dynamics is torpid: its inverse spectral gap is $O(1)$ when $\lambda \in (\lambda_c^{(k+1)},\lambda_c^{(k)})$ whereas it is $\exp[\Theta(n)]$ at each $\lambda_c^{(k)}$ for each $k\leq K_\beta$, due to a coexistence of rigid phases at heights $k+1$ and $k$. Our focus is understanding (a) the onset of metastability as $\lambda_n\uparrow\lambda_c^{(k)}$; and (b) the effect of an unbounded number of layers, as we remove the restriction $k\le K_\beta$, and even allow for $\lambda_n\to 0$ towards the $\lambda = 0$ case which has $O(\log n)$ layers and was studied by Caputo et al. (2014). We show that for any $k$, possibly growing with $n$, the inverse gap is $\exp[\tilde\Theta(1/|\lambda_n-\lambda_c^{(k)}|)]$ as $\lambda\uparrow \lambda_c^{(k)}$ up to distance $n^{-1+o(1)}$ from this critical point, due to a metastable layer at height $k$ on the way to forming the desired layer at height $k+1$. By taking $\lambda_n = n^{-\alpha}$ (corresponding to $k_n\asymp \log n$), this also interpolates down to the behavior of the dynamics when $\lambda =0$. We complement this by extending the fast mixing to all $\lambda$ uniformly bounded away from $(\lambda_c^{(k)})_{k=0}^\infty$. Together, these results provide a sharp understanding of the predicted infinite sequence of dynamical phase transitions governed by the layering phenomenon., Comment: 49 pages, 3 figures

Let $\mathcal{O}$ be a maximal order in the quaternion algebra over $\mathbb{Q}$ ramified at $p$ and $\infty$. We prove two theorems that allow us to recover the structure of $\mathcal{O}$ from limited information. The first says that for any infinite set $S$ of integers coprime to $p$, $\mathcal{O}$ is spanned as a $\mathbb{Z}$-module by elements with norm in $S$. The second says that $\mathcal{O}$ is determined up to isomorphism by its theta function., Comment: 29 pages, 3 figures

Malnutrition and dehydration are strongly associated with increased cognitive and functional decline in people living with dementia (PLWD), as well as an increased rate of hospitalisations in comparison to their healthy counterparts. Extreme changes in eating and drinking behaviours can often lead to malnutrition and dehydration, accelerating the progression of cognitive and functional decline and resulting in a marked reduction in quality of life. Unfortunately, there are currently no established methods by which to objectively detect such changes. Here, we present the findings of an extensive quantitative analysis conducted on in-home monitoring data collected from 73 households of PLWD using Internet of Things technologies. The Coronavirus 2019 (COVID-19) pandemic has previously been shown to have dramatically altered the behavioural habits, particularly the eating and drinking habits, of PLWD. Using the COVID-19 pandemic as a natural experiment, we conducted linear mixed-effects modelling to examine changes in mean kitchen activity within a subset of 21 households of PLWD that were continuously monitored for 499 days. We report an observable increase in day-time kitchen activity and a significant decrease in night-time kitchen activity (t(147) = -2.90, p < 0.001). We further propose a novel analytical approach to detecting changes in behaviours of PLWD using Markov modelling applied to remote monitoring data as a proxy for behaviours that cannot be directly measured. Together, these results pave the way to introduce improvements into the monitoring of PLWD in naturalistic settings and for shifting from reactive to proactive care., Comment: 12 pages, 7 figures, journal

Machine-learning models are known to be vulnerable to evasion attacks that perturb model inputs to induce misclassifications. In this work, we identify real-world scenarios where the true threat cannot be assessed accurately by existing attacks. Specifically, we find that conventional metrics measuring targeted and untargeted robustness do not appropriately reflect a model's ability to withstand attacks from one set of source classes to another set of target classes. To address the shortcomings of existing methods, we formally define a new metric, termed group-based robustness, that complements existing metrics and is better-suited for evaluating model performance in certain attack scenarios. We show empirically that group-based robustness allows us to distinguish between models' vulnerability against specific threat models in situations where traditional robustness metrics do not apply. Moreover, to measure group-based robustness efficiently and accurately, we 1) propose two loss functions and 2) identify three new attack strategies. We show empirically that with comparable success rates, finding evasive samples using our new loss functions saves computation by a factor as large as the number of targeted classes, and finding evasive samples using our new attack strategies saves time by up to 99\% compared to brute-force search methods. Finally, we propose a defense method that increases group-based robustness by up to 3.52$\times$.

The ultrafast formation of strongly bound excitons in two-dimensional semiconductors provide a rich platform for studying fundamental physics as well as developing novel optoelectronic technologies. While extensive research has explored the excitonic coherence, many-body interactions, and nonlinear optical properties, the potential to study these phenomena by directly controlling their coherent polarization dynamics has not been fully realized. In this work, we use a sub-10fs pulse shaper to study how temporal control of coherent exciton polarization affects the generation of four-wave mixing in monolayer WSe2 under ambient conditions. By tailoring multiphoton pathway interference, we tune the nonlinear response from destructive to constructive interference, resulting in a 2.6-fold enhancement over the four-wave mixing generated by a transform-limited pulse. This demonstrates a general method for nonlinear enhancement by shaping the pulse to counteract the temporal dispersion experienced during resonant light-matter interactions. Our method allows us to excite both 1s and 2s states, showcasing a selective control over the resonant state that produces nonlinearity. By comparing our results with theory, we find that exciton-exciton interactions dominate the nonlinear response, rather than Pauli blocking. This capability to manipulate exciton polarization dynamics in atomically thin crystals lays the groundwork for exploring a wide range of resonant phenomena in condensed matter systems and opens up new possibilities for precise optical control in advanced optoelectronic devices.

The interface between the plus and minus phases in the low temperature 3D Ising model has been intensely studied since Dobrushin's pioneering works in the early 1970's established its rigidity. Advances in the last decade yielded the tightness of the maximum of the interface of this Ising model on the cylinder of side length $n$, around a mean that is asymptotically $c\log n$ for an explicit $c$ (temperature dependent). In this work, we establish analogous results for the 3D Potts and random cluster (FK) models. Compared to 3D Ising, the Potts model and its lack of monotonicity form obstacles for existing methods, calling for new proof ideas, while its interfaces (and associated extrema) exhibit richer behavior. We show that the maxima and minima of the interface bounding the blue component in the 3D Potts interface, and those of the interface bounding the bottom component in the 3D FK model, are governed by 4 different large deviation rates, whence the corresponding global extrema feature 4 distinct constants $c$ as above. Due to the above obstacles, our methods are initially only applicable to 1 of these 4 interface extrema, and additional ideas are needed to recover the other 3 rates given the behavior of the first one., Comment: 60 pages; 8 figures

In this study, we use recent stance detection methods to study the stance (for, against or neutral) of statements in official information booklets for voters. Our main goal is to answer the fundamental question: are topics to be voted on presented in a neutral way? To this end, we first train and compare several models for stance detection on a large dataset about Swiss politics. We find that fine-tuning an M-BERT model leads to the best accuracy. We then use our best model to analyze the stance of utterances extracted from the Swiss federal voting booklet concerning the Swiss popular votes of September 2022, which is the main goal of this project. We evaluated the models in both a multilingual as well as a monolingual context for German, French, and Italian. Our analysis shows that some issues are heavily favored while others are more balanced, and that the results are largely consistent across languages. Our findings have implications for the editorial process of future voting booklets and the design of better automated systems for analyzing political discourse. The data and code accompanying this paper are available at https://github.com/ZurichNLP/voting-booklet-bias., Comment: 10 pages (including abstract and appendix), 5 figures, Keywords: stance detection, natural language processing, political analysis

We test similarity-based word alignment models (SimAlign and awesome-align) in combination with word embeddings from mBERT and XLM-R on parallel sentences in German and Romansh. Since Romansh is an unseen language, we are dealing with a zero-shot setting. Using embeddings from mBERT, both models reach an alignment error rate of 0.22, which outperforms fast_align, a statistical model, and is on par with similarity-based word alignment for seen languages. We interpret these results as evidence that mBERT contains information that can be meaningful and applicable to Romansh. To evaluate performance, we also present a new trilingual corpus, which we call the DERMIT (DE-RM-IT) corpus, containing press releases made by the Canton of Grisons in German, Romansh and Italian in the past 25 years. The corpus contains 4 547 parallel documents and approximately 100 000 sentence pairs in each language combination. We additionally present a gold standard for German-Romansh word alignment. The data is available at https://github.com/eyldlv/DERMIT-Corpus.

One-hot maps are commonly used in the AI domain. Unsurprisingly, they can also bring great benefits to ML-based algorithms such as decision trees that run under Homomorphic Encryption (HE), specifically CKKS. Prior studies in this domain used these maps but assumed that the client encrypts them. Here, we consider different tradeoffs that may affect the client's decision on how to pack and store these maps. We suggest several conversion algorithms when working with encrypted data and report their costs. Our goal is to equip the ML over HE designer with the data it needs for implementing encrypted one-hot maps.