We introduce a deep learning model for speech denoising, long-standing challenge in audio analysis arising numerous applications. Our approach is based on key observation about human speech: there often short pause between each sentence or word. In recorded signal, those pauses series of time periods during which only noise present. leverage these incidental silent intervals to learn automatic denoising given mono-channel audio. Detected over expose not just pure but its time-varying...

Reproducing realistic collective behaviors presents a captivating yet formidable challenge. Traditional rule-based methods rely on hand-crafted principles, limiting motion diversity and realism in generated behaviors. Recent imitation learning learn from data but often require ground-truth trajectories struggle with authenticity, especially high-density groups erratic movements. In this paper, we present scalable approach, Collective Behavior Imitation Learning (CBIL), for fish schooling...

We present a bio-inspired fish simulation platform, which we call "Foids", to generate realistic synthetic datasets for an use in computer vision algorithm training. This is first-of-its-kind dataset platform fish, generates all the 3D scenes just with simulation. One of major challenges deep learning based preparation annotated dataset. It already hard collect good quality video enough variations; moreover, it painful process annotate sufficiently large frame by frame. especially true when...

The purpose of the article is development an interface which closely adapts to individual. By quantifying frustration as human manipulates machines from a biomedical signal and making it into teaching signals machine learning (ML), we aim at system in human. authors extract characteristic vector whether examinee discomfort or not electroencephalograms (EEG) electromyograms (EMG). An artificial neural network (ANN) employed vector. For learning, reinforcement used rewards are extracted...

In this research we compare general harmonic wavelet transforms (GHWT), constant Q (CQT) and the Cone kernel time-frequency distribution (CKTFD) for analysis of musical signals. The first two consist a series band pass filters that have (quality), each which correspond to semitone interval (or better) in scale. CKTFD is not type transform but belongs more class bilinear distributions with special property reducing (usually) undesirable interference terms common these types distributions....

免疫アルゴリズム(IA)を用いて、メタノール水蒸気改質反応用のCu/Al2O3触媒組成の最適化を行った。これまで触媒開発に用いられてきた遺伝的アルゴリズム(GA)と比較して、IAは局所解に陥りにくく広範囲を探索するのに有効である。人工ニューラルネットワーク(ANN)により触媒反応を近似した関数に対して、IAとGAを適用した。ANNにより近似した二つのピークを持つ関数を探索した結果、GAでは極大値付近の組成のみを提示した一方、IAでは二つのピーク近傍の組成を提示した。多成分触媒を対象とした開発の初期段階では、ANNによる近似関数は複数のピークを持つことが予想される。したがって、広範囲を探索し潜在的な候補も提示できるIAとANNを組み合わせた手法が有効であると考えられる。

Our approach is to achieve the cooperative behavior of autonomous decentralized agents constructed with Q-Learning, which a type reinforcement learning. The piano mover's problem employed. We propose multi agent architecture that has an external and internal agents. Internal are heterogeneous they can communicate each other. movement depends on composition actions According learning its own shape by agents, it expected avoid obstacles. simulate our method two-dimensional continuous world....

We propose a new method called "Dark Side Ternary Stars (DSTS)", which is expected to allow emotional personal agents interact with users. DSTS focuses on proper nouns adapt preference In this paper, we report only "GAGPL" in the DSTS. GAGPL for calculating positive or negative value per sentence an individual user. aim represent famous Japanese proverb "The misery of others as sweet honey". Our proposed consists Genetic Algorithms (GA) and Programming Like (GPL) algorithms. GA learns values...

Realtime speech denoising has been long studied. Almost all existing methods process the incoming data stream using a sliding window of fixed-size. Yet, we show that use fixed-size may lead to an accumulating lag, especially in presence other background computing processes occupy CPU resources. In response, propose new strategy and lightweight neural network leverage it. Our experiments proposed approach achieves quality on par with stateof-the-art realtime models. More importantly, our is...

We present a novel real-time tool for sewing together 2D patterns, enabling quick assembly of visually plausible, interactively animated garments virtual characters. The process is assisted by ad-hoc visual hints and allows designers to import patterns from any CAD-tool, connect them using seams around 3D character with body type, assess the overall quality during animation. cloth numerically simulated including robust modeling contact itself character's body. Overall, our fast prototyping...

In recent years, it became clear that RNA possess a mechanism shaping functional forms by themselves. We focus attention on riboswitch mRNA. run simulation of folding into conformation from chain. propose boid extended model the rules considered bonding powers affecting each nucleotide. Furthermore, we consider self-organization mRNA chain structures.

Personalized systems are required in many domains. However, gathering training data for personalization from individuals, as is necessary with deep learning, a difficult and time-consuming task. With our proposed method, less or no to adapt individuals' preferences, even when they shift over time. We introduce potential field based method "Dark Side Ternary Stars" which has three components, GAGPL, Wordoids, EGO. In this paper, we focus on two of them, "Wordoids", adopt extends Boids...