Search

State-of-the art Text-To-Music (TTM) generative AI models are large and require desktop or server class compute, making them infeasible for deployment on mobile phones. This paper presents an analysis of trade-offs between model compression and generation performance of TTM models. We study compression through knowledge distillation and specific modifications that enable applicability over the various components of the TTM model (encoder, generative model and the decoder). Leveraging these methods we create TinyTTM (89.2M params) that achieves a FAD of 3.66 and KL of 1.32 on MusicBench dataset, better than MusicGen-Small (557.6M params) but not lower than MusicGen-small fine-tuned on MusicBench., Comment: Proceedings of INTERSPEECH 2024

This paper is the initial part of a comprehensive study of spacetimes that admit the canonical forms of Killing tensor in General Relativity. The general scope of the study is to derive either new exact solutions of Einstein's equations that exhibit hidden symmetries or to identify the hidden symmetries in already known spacetimes that may emerge during the resolution process. In this preliminary paper, we first introduce the canonical forms of Killing tensor, based on a geometrical approach to classify the canonical forms of symmetric 2-rank tensors, as postulated by R. V. Churchill. Subsequently, the derived integrability conditions of the canonical forms serve as additional equations transforming the under-determined system of equations, comprising of Einstein's Field Equations and the Bianchi Identities (in vacuum with {\Lambda}), into an over-determined one. Using a null rotation around the null tetrad frame we manage to simplify the system of equations to the point where the geometric characterization (Petrov Classification) of the extracted solutions can be performed and their null congruences can be characterized geometrically. Therein, we obtain multiple special algebraic solutions according to the Petrov classification (D, III, N, O) where some of them appeared to be new. The latter becomes possible since our analysis is embodied with the usage of the Newman-Penrose formalism of null tetrads., Comment: 27 pages

Causal inference has gained much popularity in recent years, with interests ranging from academic, to industrial, to educational, and all in between. Concurrently, the study and usage of neural networks has also grown profoundly (albeit at a far faster rate). What we aim to do in this blog write-up is demonstrate a Neural Network causal inference architecture. We develop a fully connected neural network implementation of the popular Bayesian Causal Forest algorithm, a state of the art tree based method for estimating heterogeneous treatment effects. We compare our implementation to existing neural network causal inference methodologies, showing improvements in performance in simulation settings. We apply our method to a dataset examining the effect of stress on sleep.

Unsupervised learning has gained prominence in the big data era, offering a means to extract valuable insights from unlabeled datasets. Deep clustering has emerged as an important unsupervised category, aiming to exploit the non-linear mapping capabilities of neural networks in order to enhance clustering performance. The majority of deep clustering literature focuses on minimizing the inner-cluster variability in some embedded space while keeping the learned representation consistent with the original high-dimensional dataset. In this work, we propose soft silhoutte, a probabilistic formulation of the silhouette coefficient. Soft silhouette rewards compact and distinctly separated clustering solutions like the conventional silhouette coefficient. When optimized within a deep clustering framework, soft silhouette guides the learned representations towards forming compact and well-separated clusters. In addition, we introduce an autoencoder-based deep learning architecture that is suitable for optimizing the soft silhouette objective function. The proposed deep clustering method has been tested and compared with several well-studied deep clustering methods on various benchmark datasets, yielding very satisfactory clustering results.

Large Language Models (LLMs) exhibit impressive performance on a range of NLP tasks, due to the general-purpose linguistic knowledge acquired during pretraining. Existing model interpretability research (Tenney et al., 2019) suggests that a linguistic hierarchy emerges in the LLM layers, with lower layers better suited to solving syntactic tasks and higher layers employed for semantic processing. Yet, little is known about how encodings of different linguistic phenomena interact within the models and to what extent processing of linguistically-related categories relies on the same, shared model representations. In this paper, we propose a framework for testing the joint encoding of linguistic categories in LLMs. Focusing on syntax, we find evidence of joint encoding both at the same (related part-of-speech (POS) classes) and different (POS classes and related syntactic dependency relations) levels of linguistic hierarchy. Our cross-lingual experiments show that the same patterns hold across languages in multilingual LLMs., Comment: Accepted in EMNLP Findings 2023

The study of the Canonical Forms of the Killing Tensor concerns the simultaneous resolving of the Integrability Conditions of the Killing Tensor along with the Einstein's Field Equations employing the framework of Newman-Penrose Formalism. We present all the Petrov Types admitting the 2nd and 3rd Canonical Forms of Killing Tensor in Vacuum in the frame of General Theory of Relativity. During the investigation of the Type D solution of 2nd Canonical form of the Killing Tensor the Carter's Case [D] solution in Vacuum emerged.

Ambiguous questions are a challenge for Question Answering models, as they require answers that cover multiple interpretations of the original query. To this end, these models are required to generate long-form answers that often combine conflicting pieces of information. Although recent advances in the field have shown strong capabilities in generating fluent responses, certain research questions remain unanswered. Does model/data scaling improve the answers' quality? Do automated metrics align with human judgment? To what extent do these models ground their answers in evidence? In this study, we aim to thoroughly investigate these aspects, and provide valuable insights into the limitations of the current approaches. To aid in reproducibility and further extension of our work, we open-source our code at https://github.com/din0s/ambig_lfqa., Comment: Accepted in the Findings of ACL 2023

The lack of interpretability of the Vision Transformer may hinder its use in critical real-world applications despite its effectiveness. To overcome this issue, we propose a post-hoc interpretability method called VISION DIFFMASK, which uses the activations of the model's hidden layers to predict the relevant parts of the input that contribute to its final predictions. Our approach uses a gating mechanism to identify the minimal subset of the original input that preserves the predicted distribution over classes. We demonstrate the faithfulness of our method, by introducing a faithfulness task, and comparing it to other state-of-the-art attribution methods on CIFAR-10 and ImageNet-1K, achieving compelling results. To aid reproducibility and further extension of our work, we open source our implementation: https://github.com/AngelosNal/Vision-DiffMask, Comment: Accepted in the XAI4CV Workshop at CVPR 2023

Studying under-represented music traditions under the MIR scope is crucial, not only for developing novel analysis tools, but also for unveiling musical functions that might prove useful in studying world musics. This paper presents a dataset for Greek Traditional and Folk music that includes 1570 pieces, summing in around 80 hours of data. The dataset incorporates YouTube timestamped links for retrieving audio and video, along with rich metadata information with regards to instrumentation, geography and genre, among others. The content has been collected from a Greek documentary series that is available online, where academics present music traditions of Greece with live music and dance performance during the show, along with discussions about social, cultural and musicological aspects of the presented music. Therefore, this procedure has resulted in a significant wealth of descriptions regarding a variety of aspects, such as musical genre, places of origin and musical instruments. In addition, the audio recordings were performed under strict production-level specifications, in terms of recording equipment, leading to very clean and homogeneous audio content. In this work, apart from presenting the dataset in detail, we propose a baseline deep-learning classification approach to recognize the involved musicological attributes. The dataset, the baseline classification methods and the models are provided in public repositories. Future directions for further refining the dataset are also discussed.

The field of automatic music composition has seen great progress in recent years, specifically with the invention of transformer-based architectures. When using any deep learning model which considers music as a sequence of events with multiple complex dependencies, the selection of a proper data representation is crucial. In this paper, we tackle the task of conditional drums generation using a novel data encoding scheme inspired by the Compound Word representation, a tokenization process of sequential data. Therefore, we present a sequence-to-sequence architecture where a Bidirectional Long short-term memory (BiLSTM) Encoder receives information about the conditioning parameters (i.e., accompanying tracks and musical attributes), while a Transformer-based Decoder with relative global attention produces the generated drum sequences. We conducted experiments to thoroughly compare the effectiveness of our method to several baselines. Quantitative evaluation shows that our model is able to generate drums sequences that have similar statistical distributions and characteristics to the training corpus. These features include syncopation, compression ratio, and symmetry among others. We also verified, through a listening test, that generated drum sequences sound pleasant, natural and coherent while they "groove" with the given accompaniment., Comment: Accepted for the 11th International Conference on Artificial Intelligence in Music, Sound, Art and Design (EvoMUSART), 2022

This paper is an overview of the Machine Learning Operations (MLOps) area. Our aim is to define the operation and the components of such systems by highlighting the current problems and trends. In this context, we present the different tools and their usefulness in order to provide the corresponding guidelines. Moreover, the connection between MLOps and AutoML (Automated Machine Learning) is identified and how this combination could work is proposed., Comment: 8 pages, figure 5, table 4

Over the last decades, images have become an important source of information in many domains, thus their high quality has become necessary to acquire better information. One of the important issues that arise is image denoising, which means recovering a signal from inaccurately and/or partially measured samples. This interpretation is highly correlated to the compressive sensing theory, which is a revolutionary technology and implies that if a signal is sparse then the original signal can be obtained from a few measured values, which are much less, than the ones suggested by other used theories like Shannon's sampling theories. A strong factor in Compressive Sensing (CS) theory to achieve the sparsest solution and the noise removal from the corrupted image is the selection of the basis dictionary. In this paper, Discrete Cosine Transform (DCT) and moment transform (Tchebichef, Krawtchouk) are compared in order to achieve image denoising of Gaussian additive white noise based on compressive sensing and sparse approximation theory. The experimental results revealed that the basis dictionaries constructed by the moment transform perform competitively to the traditional DCT. The latter transform shows a higher PSNR of 30.82 dB and the same 0.91 SSIM value as the Tchebichef transform. Moreover, from the sparsity point of view, Krawtchouk moments provide approximately 20-30% more sparse results than DCT., Comment: 12 pages, 13 figures

It is widely speculated that auditors' public forecasts of bankruptcy are, at least in part, self-fulfilling prophecies in the sense that they might actually cause bankruptcies that would not have otherwise occurred. This conjecture is hard to prove, however, because the strong association between bankruptcies and bankruptcy forecasts could simply indicate that auditors are skillful forecasters with unique access to highly predictive covariates. In this paper, we investigate the causal effect of bankruptcy forecasts on bankruptcy using nonparametric sensitivity analysis. We contrast our analysis with two alternative approaches: a linear bivariate probit model with an endogenous regressor, and a recently developed bound on risk ratios called E-values. Additionally, our machine learning approach incorporates a monotonicity constraint corresponding to the assumption that bankruptcy forecasts do not make bankruptcies less likely. Finally, a tree-based posterior summary of the treatment effect estimates allows us to explore which observable firm characteristics moderate the inducement effect., Comment: 26 pages, 13 figures

The capability to automatically detect human stress can benefit artificial intelligent agents involved in affective computing and human-computer interaction. Stress and emotion are both human affective states, and stress has proven to have important implications on the regulation and expression of emotion. Although a series of methods have been established for multimodal stress detection, limited steps have been taken to explore the underlying inter-dependence between stress and emotion. In this work, we investigate the value of emotion recognition as an auxiliary task to improve stress detection. We propose MUSER -- a transformer-based model architecture and a novel multi-task learning algorithm with speed-based dynamic sampling strategy. Evaluations on the Multimodal Stressed Emotion (MuSE) dataset show that our model is effective for stress detection with both internal and external auxiliary tasks, and achieves state-of-the-art results., Comment: NAACL 2021 accepted

Natural Language Processing (NLP) has evolved significantly over the last decade. This paper highlights the most important milestones of this period while trying to pinpoint the contribution of each individual model and algorithm to the overall progress. Furthermore, it focuses on issues still remaining to be solved, emphasizing the groundbreaking proposals of Transformers, BERT, and all the similar attention-based models., Comment: 13 pages, 5 figures

This paper highlights the trends in the field of predictive maintenance with the use of machine learning. With the continuous development of the Fourth Industrial Revolution, through IoT, the technologies that use artificial intelligence are evolving. As a result, industries have been using these technologies to optimize their production. Through scientific research conducted for this paper, conclusions were drawn about the trends in Predictive Maintenance applications with the use of machine learning bridging Artificial Intelligence and IoT. These trends are related to the types of industries in which Predictive Maintenance was applied, the models of artificial intelligence were implemented, mainly of machine learning and the types of sensors that are applied through the IoT to the applications. Six sectors were presented and the production sector was dominant as it accounted for 54.54% of total publications. In terms of artificial intelligence models, the most prevalent among ten were the Artificial Neural Networks, Support Vector Machine and Random Forest with 27.84%, 17.72% and 13.92% respectively. Finally, twelve categories of sensors emerged, of which the most widely used were the sensors of temperature and vibration with percentages of 60.71% and 46.42% correspondingly., Comment: 7 pages, 7 figures

This paper deals with the identification of machines in a smart city environment. The concept of machine biometrics is proposed in this work for the first time, as a way to authenticate machine identities interacting with humans in everyday life. This definition is imposed in modern years where autonomous vehicles, social robots, etc. are considered active members of contemporary societies. In this context, the case of car identification from the engine behavioral biometrics is examined. For this purpose, 22 sound features were extracted and their discrimination capabilities were tested in combination with 9 different machine learning classifiers, towards identifying 5 car manufacturers. The experimental results revealed the ability of the proposed biometrics to identify cars with high accuracy up to 98% for the case of the Multilayer Perceptron (MLP) neural network model., Comment: 5 pages, 4 figures

Nowadays, an important topic that is considered a lot is how to integrate Machine Learning(ML) to cloud resources management. In this study, our goal is to explore the most important cloud resources management issues that have been combined with ML and which present many promising results. To accomplish this, we used chronological charts based on some keywords that we considered important and tried to answer the question: is ML suitable for resources management problems in the cloud? Furthermore, a short discussion takes place on the data that are available and the open challenges on it. A big collection of researches is used to make sensible comparisons between the ML techniques that are used in the different kinds of cloud resources management fields and we propose the most suitable ML model for each field. 1, Comment: 13 pages, 3 figures, to be published in proceedings of International Conference on Expert Clouds and Applications (ICOECA 2021)

In this paper, a performance evaluation of well-known deep learning models in gait recognition is presented. For this purpose, the transfer learning scheme is adopted to pre-trained models in order to fit the models to the CASIA-B dataset for solving a gait recognition task. In this context, 18 popular Convolutional Neural Networks (CNNs), were re-trained using Gait Energy Images (GEIs) of CASIA-B containing almost 14000 images of 124 classes under various conditions, and their performance was studied in terms of accuracy. Moreover, the performance of the studied models is managed to be explained by examining the parts of the images being considered by the models towards providing their decisions. The experimental results are very promising since almost all the models achieved a high accuracy of over 90%, which is robust to the increasing number of classes. Furthermore, an important outcome of this study is the fact that a recognition problem can be effectively solved by using CNNs pre-trained to different problems, thus eliminating the need for customized model design., Comment: 6 pages, 15 figures, to be published in proceedings of the 24th Pan-Hellenic Conference on Informatics (PCI)