There is a growing trend to teach playing an instrument such as a piano at home using an automated system. A key component of such systems is the ability to rate performance of the learner in order to provide feedback and select appropriate exercises. In this study, we expand on previous works that have developed automatic evaluation systems for an overall grade by also providing predictions for specific aspects of performance: pitch, rhythm, tempo, and articulation & dynamics, as well as scheduling what is an appropriate next task. We describe how a set of salient features is extracted by comparing MIDI performance data of three piano players to an ideal performance, how the features used for evaluation are selected, and evaluate using linear regression how well the selected features are able to predict the mean scores given by a group of domain experts (piano teachers). Relatively good R2 scores (0.54 to 0.68) are achieved using a small number of features (2 - 4). Such automatic evaluation of different aspects of performance can be used as a part of an automatic learning system, and to help provide learners with detailed feedback on their performance.

Abstract A major property of flash memory is that a 0-bit can be changed into a 1-bit, but the symmetric task of switching from a 1-bit to a zero may only be performed in blocks and is therefore often prohibited. This led to the development of rewriting codes using the same storage space more than once, subject to the constraint that 0-bits can be changed into 1-bits, but not vice versa. Context sensitive rewriting codes extend this idea by also incorporating information gathered from surrounding bits. Several new context sensitive rewriting codes are presented and analyzed, some of which are better than the state of the art for sparse input. Empirical simulations show a good match with the theoretical results.

A binary Write Once Memory (WOM) device is a storage mechanism in which a 0-bit can be overwritten much more easily than a 1-bit. A famous example is the flash memory technology, where 0→1 transitions are allowed, but 1 → 0 transitions require a costly erase procedure and are therefore prohibited. A WOM code is a coding scheme that permits multiple writes to the WOM without violating the WOM rule. The properties of WOM attracted attention even before flash memory was invented. Rivest and Shamir [2] proposed an elegant WOM code that uses 3 bits to write two rounds of any combination of 2 bits. As alternative, context sensitive rewriting codes have been considered [1], in which the new information stored in the second round utilizes certain portions of the output of the first round for unambiguous reuse. Some of these encodings are based on Fibonacci Codes, whose primary relevant property is that they contain no adjacent 1-bits, except as suffixes of all of their codewords.

Rewriting codes for flash memory enable the multiple usage of the same storage space, under the constraint that 0-bits can be changed into 1s, but not vice versa. Context sensitive rewriting codes extend this idea by incorporating also information gathered from surrounding bits. Several new and better context sensitive rewriting codes are presented and analyzed. Empirical simulations show a good match with the theoretical results.

Distributed learning is central for large-scale training of deep-learning models. However, it is exposed to a security threat in which Byzantine participants can interrupt or control the learning process. Previous attack models assume that the rogue participants (a) are omniscient (know the data of all other participants), and (b) introduce large changes to the parameters. Accordingly, most defense mechanisms make a similar assumption and attempt to use statistically robust methods to identify and discard values whose reported gradients are far from the population mean. We observe that if the empirical variance between the gradients of workers is high enough, an attacker could take advantage of this and launch a non-omniscient attack that operates within the population variance. We show that the variance is indeed high enough even for simple datasets such as MNIST, allowing an attack that is not only undetected by existing defenses, but also uses their power against them, causing those defense mechanisms to consistently select the byzantine workers while discarding legitimate ones. We demonstrate our attack method works not only for preventing convergence but also for repurposing of the model behavior (``backdooring''). We show that less than 25\% of colluding workers are sufficient to degrade the accuracy of models trained on MNIST, CIFAR10 and CIFAR100 by 50\%, as well as to introduce backdoors without hurting the accuracy for MNIST and CIFAR10 datasets, but with a degradation for CIFAR100.

Hierarchical Clustering is widely used in Machine Learning and Data Mining. It stores bit-vectors in the nodes of a k-ary tree, usually without trying to compress them. We suggest a double usage of the extorting operations defining the Hamming distance used in the clustering process, extending it also to be used to transform the vector in one node into a more compressible form, as a function of the vector in the parent node. Compression is then achieved by run-length encoding, followed by optional Huffman coding, and we show how the compressed file may be processed directly, without decompression.

In previous work we have suggested a data structure based on pruning a Huffman shaped Wavelet tree according to the underlying skeleton Huffman tree. This pruned Wavelet tree was especially designed to support faster random access and save memory storage, at the price of less effective rank and select operations, as compared to the original Huffman shaped Wavelet tree. In this paper we improve the pruning procedure and give empirical evidence that when memory storage is of main concern, our suggested data structure outperforms other direct access techniques such as those due to Külekci, DACs and sampling, with a slowdown as compared to DACs and fixed length encoding.