Automatic singing quality assessment is a challenging task due to its subjective nature, as well as limitations stemming from the scarcity of human annotations and the poor generalization to unfamiliar singing expressions. In this paper, we incorporate perturbed augmented data strategies into the training of a singing quality assessment (SQA) network without additional annotations. In addition, we also introduce the training of a classifier to enhance the network’s latent space in response to pitch, tempo, and timbre variation. The experiments demonstrated that our proposed SQA network outperformed the baseline with a 6.67% improvement in terms of the Pearson correlation coefficient.

The burgeoning field of text-to-music generation models has shown great promise in their ability to generate high-quality music aligned with users' textual descriptions. These models effectively capture abstract/global musical features such as style and mood. However, they often inadequately produce the precise rendering of critical music loop attributes, including melody, rhythms, and instrumentation, which are essential for modern music loop production. To overcome this limitation, this paper proposed a Loops Transformer and a Multi-Stage Cross Attention mechanism that enable a cohesive integration of textual and MIDI input specifications. Additionally, a novel Instrument-Aware Reinforcement Learning technique was introduced to ensure the correct adoption of instrumentation. We demonstrated that the proposed model can generate music loops that simultaneously satisfy the conditions specified by both natural language texts and MIDI input, ensuring coherence between the two modalities. We also showed that our model outperformed the state-of-the-art baseline model, MusicGen, in both objective metrics (by lowering the FAD score by 1.3, indicating superior quality with lower scores, and by improving the Normalized Dynamic Time Warping Distance with given melodies by 12%) and subjective metrics (by +2.56% in OVL, +5.42% in REL, and +7.74% in Loop Consistency). These improvements highlight our model's capability to produce musically coherent loops that satisfy the complex requirements of contemporary music production, representing a notable advancement in the field. Generated music loop samples can be explored at: https://loopstransformer.netlify.app/.

We adopt the recent cycle consistent generative adversarial network (MaskCycleGAN-VC) that allows converting a specific singing technique using only a few articulations of singing voice as examples. Since it is often prone to fail to preserve the content information of the singing voice due to distortion and noise during the conversion, a self-supervised learning module is proposed as the basic framework to enforce content consistency without additional annotations. We evaluate the proposed methods on three datasets that were commonly used in pop songs which involve singing techniques in terms of breathy voice, vibrato, and vocal fry. Experiments showed that our proposed methods outperform the baseline in terms of audio quality and content preservation, including melody and singer’s timbral identity, without affecting the perception of singing techniques.1

Singing techniques are important skills for a professional vocal performance that usually involves dedicated fluctuations of timbre, pitch, duration, and loudness, etc. To recognize types of singing techniques can be quite challenging because 1) the time-frequency features in singing are highly dynamic that may appear in a long range of audio signals; 2) different singing techniques such as vibrato and trill tend to have similar features in the locality; 3) The distribution of singing technique dataset suffers from the long-tailed issue. To man-age these problems, we proposed a novel Radial Attention Transformer (RAT) with a Radial Attention (RA) Module that can capture the fine-grained local features as well as the long range inter-dependency of audio features. The experiment results showed that the proposed method, RAT with Adaptive Logit Adjustment (ALA) Loss significantly outperformed pre-vious state-of-the-art models (Convolutional Neural Networks and Deformable CNN), on the recognition tasks of singing technique categories.

The scarcity of expansive datasets for singing quality assessment makes the utilization of complex deep learning methods a considerable challenge. This research presents a method to improve the singing quality prediction based on the feedback from subjective human perception opinion that is learned by the transfer learning methods of self-supervised learning (SSL) speech models. In combination with the CRNN_PH model as the baseline model, the SSL models are integrated into two distinct major architectures: one directly draws features from the pre-trained SSL model (CRNN_PH+SSL), and the other employs the weighted sum (WS) of the output features from different transformer blocks in the SSL model (CRNN_PH+SSL_WS). We conducted comparative experiments on pre-trained SSL models, five on wav2vec 2.0 (W2V2) and two on HuBERT, which were trained over various datasets. It turns out that CRNN_PH+W2V2_base_WS is improved the most on singing quality score prediction that is closely aligning with subjective human perceptions in terms of correlation coefficients and MSE with respect to the ground truth.

In the audio source separation topic, most researchers based on deep learning methods ignored higher-frequency signals due to lack of efficient data compression method. We propose a new model named OvertoneNet (OveNet) that adopts two novel concepts, frequency 1x1 convolution layers, and complex-spectrogram channels, to handle the 44.1k audio signals (Hi-Res audio signals) containing full overtones. The result shows that OveNet performs well in both objective and subjective evaluation on interference using limited training data from SiSEC2018.

With the information of drug, disease phenotype and protein interactions accumulating rapidly, to investigate the relationships between drugs and diseases is a critical importance issue. Until recently, few studies attempt to discover drug-disease associations on a network basis. We integrate drug and phenotype information and protein interaction network together and apply a network propagation approach to infer and evaluate the likelihood of the probability between drug and disease based on gene expression profile. In the experiments, we adopt prostate cancer as our test data. We validate our results to the manually curated associations in Comparative Toxicogenomics Database. Our experimental studies show that our proposed method obtains high specificity and sensitivity (AUC=0.98) and clearly outperforms previous existing methods. Our proposed method discovers potential drug-disease associations that drew the attention of biologists and provides a new perspective for toxicogenomics and drug reposition evaluation.

The identification of significant disease-related genes and networks is an important issue in understanding underlying mechanisms of cells. We integrate phenotype networks, protein networks and efficiently utilize gene expression data to identify human disease networks. We use prostate cancer data as our test domain. In comparison with statistical methods such as t-test and Wilcoxon test, our method identifies more prostate cancer-related genes reported in published database and literature. Interleukin-type growth factors, Ras related oncogenes and cytokine interactions canonical pathways are found to be significantly related to prostate cancer.

High dimensionality of state representation is a major limitation for scale‐up in reinforcement learning (RL). This work derives the knowledge of complexity reduction from partial solutions and provides algorithms for automated dimension reduction in RL. We propose the cascading decomposition algorithm based on the spectral analysis on a normalized graph Laplacian to decompose a problem into several subproblems and then conduct parameter relevance analysis on each subproblem to perform dynamic state abstraction. The elimination of irrelevant parameters projects the original state space into the one with lower dimension in which some subtasks are projected onto the same shared subtasks. The framework could identify irrelevant parameters based on performed action sequences and thus relieve the problem of high dimensionality in learning process. We evaluate the framework with experiments and show that the dimension reduction approach could indeed make some infeasible problem to become learnable.

An agent-based social simulation is one way to add a story to simulated game universes within the game mechanics while preserving the autonomy of nonplay characters (NPCs). In this paper, we add a social reasoning element behind NPC actions to make their plans more story like. An AI planner is developed to combine plan search and logic inference about others' minds. An NPC agent equipped with the planner uses actions to change others' minds, and uses such mental changes to achieve its goal. We review the question of whether stories do arise from agent-based simulations by examining actual narrative segments generated by our NPC agents, and by an experimental exploration of the frequencies and lengths of narrative segments. A story facilitator, named divine intervention operator service (DIOS), makes stories happen when they are impossible via on-the-fly adjustment of character personalities.

The main idea in this paper is using combinatorial Web service market based mechanism (CWSMBM) to allow service demanders bid for more than one service at the same time. It takes quality of service into consideration in the design of market allocation mechanisms. We define the cost-benefit threshold for service providers to decide whether to accept or reject service tasks. We found that the cost-benefit threshold improves the global benefit value effectively but would sometimes cause a starvation problem. But the starvation can be a trade-off in sacrificing some service demanders to enhance the global benefit.

Adverse drug reaction (ADR) may cost a lot of unnecessary medical resources and leads to extra suffering on patients. To provide the prompt information about ADR and avoid the rate of occurrence of ADR is an important task yet to be done. The US Food and Drug Administration (FDA) provides a Adverse Event Reporting System (AERS) which contains a lot of clinical reports about ADRs. However, the biologists still do not know precisely which observed events are directly caused by drug-drug interactions. We use the probability analysis method to find the associations between a set of drugs and the symptoms for predicting the ADR and apply the decision tree to discovery the association rules between the drug-drug interactions and symptoms.

Microarray is widely used for the cancer research and identifies different expressions for specific genes. We present a computational method for constructing cancer and normal gene regulatory networks from micorarray data based on transcription factor analysis and independency test. The web service technology is used to wrap the bioinformatics toolkits of methods and databases to automatically extract the promoter regions of DNA sequences and predict the transcription factors that regulate gene expressions. After reconstructing the gene regulatory network, the network statistical measure and network motifs extract the potential genes to compare the sub-networks between the cancer and normal gene networks. We adopt the microarray datasets from Stanford microarray database of prostate cancer as a target application to evaluate the methods.

Patent invention includes various types of knowledge processing tasks such as patent document analysis, patent search, patent classification, patent valuation, to ensure the usefulness and novelty of the new patent invention. In the past, patent invention for an industry relied solely on tedious interactions among domain-specific human experts. In this paper, we propose a cooperative multi-agent and Web service platform to facilitate the invention of a new patent. The platform integrates web service technique as a standard information exchange mechanism for agents to communicate with each other in FIPA Agent Communication Language (ACL). We design a utility function in terms of cost, time and value of information service for an agent to decide whether a service should be selected in the workflow of the patent invention. The agents are implemented in JADE and we illustrate with two explicit examples in the domain of inventing a new mechanical design patent and show how the cooperative multi-agent and web service platform supports the patent invention process.

We develop a framework using ontology inference and semantic processing techniques to help biologists to extract knowledge directly from a large scale of biological literature in NCBI PubMed. The system integrated various sharable thesauri of WordNet, MeSH (Medical Subject Heading), and GO (Gene ontology) to support the automatic semantic annotation and analysis. The natural language processing and semantic processing are facilitated by the ontological inference, and the system could automatically extract the correct molecular interactions from the complex sentences in an abstract automatically. It facilitates the biologists not only to save time and efforts to construct and analyze biological pathways, but also to discover the novel molecular interactions by comparing the information extracted from the literature with that in such existing pathway database as KEGG. We evaluated the system performance based on the pathways in Apoptosis domain.

In a grid computing environment, each client has its own job represented as a workflow composed of tasks that require multiple types of computational resources to complete. Developing a mechanism that schedules these workflows to efficiently utilize limited amounts of resources in the grid is a challenging problem. This paper takes a market-oriented approach allowing the job scheduling task to be distributed among clients. In this approach, several workflow agents plan a feasible schedule for their jobs and compete in the resource market. A market broker agent is implemented to coordinate the conflicts in simultaneous access of the same resource. Experiment results show that the performance of the proposed approach surpasses those of first-come-first-serve and a variant of shortest-job-first method in terms of job completion ratio before deadline.

We propose a cooperative multi-agent platform to support the invention process based on the patent document analysis. It helps industrial knowledge managers to retrieve and analyze existing patent documents and extract structure information from patents with the aid of ontology and natural language processing techniques. It allows the invention process to be carried out through the cooperation and coordination among software agents delegated by the various domain experts in the complex industrial R&D environment. Furthermore, it integrates the patent document analysis with the inventive problem solving method known as TRIZ method that can suggest invention directions based on the heuristics or principles to resolve the contradictions among design objectives and engineering parameters. We chose the patent invention for Chemical Mechanical Polishing (CMP) as our case study. However, the platform and techniques could be extended to most cooperative invention domains.

As more and more new genome sequences were reported nowadays, analyzing the functions of a new genome sequence becomes more and more desirable and compelling. However, the determination of the functions of a genomic sequence is not an easy task. Even with several bioinformatic tools, the task is still a labor-intensive one. This is because human experts have to intervene during the processing of using these tools. For efficiency, immediacy and reduction of human labor, a system of automating the analyzing process is proposed. We take the automated determination of open reading frames of a genomic sequence as the domain tasks that involve using a number of computational tools and interpreting the results returned from the tools. A service-oriented approach is taken, in which analyzing tools are wrapped as Web services and described in semantic Web languages including OWL and OWL-S. The SARS coronavirus genomic sequence is taken as a test case for our approaches. We are in the process of building an agent-based system for automating the tasks, in which an intelligent agent is responsible for understanding purposes of the Web services by parsing the service descriptions, and carrying out the interpretation tasks according to a workflow.

We present a game theoretic coordination mechanism in a multi agent community. We assume that all the agents are rational and have the ability to communicate with each other. In our approach, agents are treated as players in a noncooperative game defined in conventional game theory. In order to make the agents behave coordinately and to avoid an undesirable state, such as Prisoners' dilemma, we introduce a trusted third party into the conventional two-player game theory. The mechanism changes the equilibrium states by altering the payoff of the game. We show how agents are able to recognize undesirable states by reasoning on a 2 by 2 payoff matrix and find a way out by communicating with a trusted third party. A communication protocol among agents and the trusted third party is constructed to achieve a negotiation for coordination.

The fuzzy ARTMAP is one of the families of neural network architectures based on ART (adaptive resonance theory) in which supervised learning can be carried out. However, it usually tends to create more categories than are actually needed. This often causes the so-called overfitting problem, where the performance of the fuzzy ARTMAP networks in the test set does not increase monotonically with additional training epochs and category creation. In order to avoid the overfitting problem, Carpenter and Tan (1993) proposed a confidence-based pruning method by eliminating those categories that were either less useful or less accurate. This paper proposes yet another alternative pruning method, which is based on the minimal description length (MDL) principle. The MDL principle can be viewed as a tradeoff between theory complexity and data prediction accuracy, given the theory. We adopted Cameron-Jones's (1992) error encoding scheme and Quinlan's (1994, 1995) modification for theory encoding to estimate the fuzzy ARTMAP theory description length. A greedy MDL search algorithm is proposed to prune the fuzzy ARTMAP categories one by one. Experiments showed that a fuzzy ARTMAP pruned with the MDL principle gave a better performance, with far fewer categories created, than the original fuzzy ARTMAP and other machine-learning systems on a number of benchmark clinical databases such as heart disease, breast cancer and diabetes databases.

The thematic knowledge can bridge the gap between semantic entities and syntactic constituents. In document understanding, the correctness and the efficiency could be improved if the thematic knowledge is available. In this paper, we propose a semi-automatic method to acquire thematic knowledge of Chinese verbs by exploiting syntactic clues. The syntactic clues, which may be collected by most existing syntactic processors, reduce the hypothesis space of the theta roles. The ambiguities may be further resolved by the evidences from a trainer. A set of heuristics based on linguistic constraints are employed to guide the ambiguity resolution process. To acquire thematic information for verbs, the argument structures of the verbs must be extracted first. A template matching method is used to extract the argument structure of verbs.

Thematic knowledge is a basis of semantic interpretation. In this paper, we propose an acquisition method to acquire thematic knowledge by exploiting syntactic clues from training sentences. The syntactic clues, which may be easily collected by most existing syntactic processors, reduce the hypothesis space of the thematic roles. The ambiguities may be further resolved by the evidences either from a trainer or from a large corpus. A set of heuristics based on linguistic constraints is employed to guide the ambiguity resolution process. When a trainer is available, the system generates new sentences whose thematic validities can be justified by the trainer. When a large corpus is available, the thematic validity may be justified by observing the sentences in the corpus. Using this way, a syntactic processor may become a thematic recognizer by simply deriving its thematic knowledge from its own syntactic knowledge.

Parsing is an important step in natural language processing. It involves tasks of searching for applicable grammatical rules which can transform natural language sentences into their corresponding parse trees. Therefore parsing can be viewed as problem solving, and language acquisition can be achieved by generalizing problem solving heuristics. In this paper we investigate how machine learning methodologies can be integrated with a Wait-And-See Parser (the problem solver) to acquire parsing-related knowledge that is needed for the parser. We call this approach parsing-driven generalization since learning (acquisition of parsing rules and classification of lexicons) is basically derived from the parsing process. Three types of generalization are reported in this paper: simple generalization, generalization by asking questions, and generalization back-propagation. Simple generalization generalizes any two parsing rules whose action parts (right-hand sides) are the same but whose condition parts (left-hand sides) have a single difference. Generalization by asking questions is triggered when a “climbing-up” move on a concept hierarchy is attempted. It is necessary for avoiding over-generalization. Generalization back-propagation propagates a confirmed generalization of some later parsing rule back to its precedent rules in a parsing sequence and thus causes them to be generalized as well. It can reduce the number of questions asked by the system. With these three types of generalization and a mechanism for maintaining lexicon classification (the domain concept hierarchy), parsing and learning can interact to utilize and acquire parsing-related knowledge. To promote the practical performance of parsing after learning, a relaxation parsing mechanism is also designed to process unseen sentences.

The major problems in parsing English conjunctions and comparatives are ambiguities of scoping and ellipsis. Scoping ambiguities occur when a parser cannot deterministically detect boundaries of constituents, while ellipsis ambiguities occur when a parser cannot deterministically detect missing components. Since simple lookahead mechanisms cannot collect adequate information to resolve these ambiguities, a parsing strategy that only employs such mechanisms will need to backtrack each time it makes incorrect assumptions. In this paper, we extend the Wait-And-See strategy to parse conjunctions and comparatives deterministically and simultaneously. Several mechanisms, such as bottom-up preparsing, suspension, and pattern matching, are implemented. The bottom-up preparsing accesses the dictionary and recognizes isolated sentence fragments which can be determined without ambiguities. The suspension, which is different from Marcus’s attention shifting, allows the parser to suspend temporally at ambiguous points and continue to parse the rest of the sentence until it obtains the necessary information to resolve the ambiguities. Pattern matching uses the concept of symmetry to detect missing components (the ellipses) in the two conjoined or compared sentence fragments.