We propose CoLyricist, an AI-assisted lyric writing tool designed to support the typical workflows of experienced lyricists and enhance their creative efficiency. While lyricists have unique processes, many follow common stages. Tools that fail to accommodate these stages challenge integration into creative practices. Existing research and tools lack sufficient understanding of these songwriting stages and their associated challenges, resulting in ineffective designs. Through a formative study involving semi-structured interviews with 10 experienced lyricists, we identified four key stages: Theme Setting, Ideation, Drafting Lyrics, and Melody Fitting. CoLyricist addresses these needs by incorporating tailored AI-driven support for each stage, optimizing the lyric writing process to be more seamless and efficient. To examine whether this workflow-aligned design also benefits those without prior experience, we conducted a user study with 16 participants, including both experienced and novice lyricists. Results showed that CoLyricist enhances the songwriting experience across skill levels. Novice users especially appreciated the Melody-Fitting feature, while experienced users valued the Ideation support.

Despite the success of Large Language Models (LLMs) on various tasks following human instructions, controlling model generation to follow strict constraints at inference time poses a persistent challenge. In this paper, we introduce Ctrl-G, a neuro-symbolic framework that enables tractable and adaptable control of LLM generation to follow logical constraints reliably. Ctrl-G combines any production-ready LLM with a Hidden Markov Model (HMM), guiding LLM outputs to adhere to logical constraints represented as deterministic finite automata. We show that Ctrl-G, when a TULU2-7B model is coupled with a 2B-parameter HMM, outperforms GPT4 in text editing: on the task of generating text insertions/continuations following logical constraints, our approach achieves over 30% higher satisfaction rate in human evaluation. When applied to medium-size language models (e.g., GPT2-large), Ctrl-G also beats its counterparts on standard benchmarks by large margins. Additionally, as a proof-of-concept study, we use Ctrl-G to assist LLM reasoning on the GSM benchmark, foreshadowing the application of Ctrl-G, as well as other constrained generation approaches, beyond traditional language generation tasks.

Photonic-crystal surface-emitting lasers (PCSELs) are attractive for a wide range of applications such as material processing and remote sensing owing to their advantage of broad-area coherent lasing supported by a 2-D photonic crystal. Recently, we proposed double-lattice photonic-crystal resonators and achieved 10-W-class high-power and high-beam-quality (namely, high-brightness) operation. To further increase the brightness, it is important to understand the lasing mode behavior of PCSELs in detail. In this article, we experimentally investigate the change in the lasing mode depending on the current injection levels and compare the results with a theoretical analysis that accounts for the change in refractive index and gain distributions due to the current injection. We observed a transition from single-mode lasing to two-mode lasing at high injection currents. We develop a technique for resolving the near-field profiles of individual lasing modes and show that these profiles coincide with theoretical predictions. A comparison of experimental and theoretical results reveals that high-order modes appear due to a bandgap confinement effect induced by a change in the refractive index in the carrier injection area. To effectively remove this confinement effect and suppress the oscillation of high-order modes, we incorporate an in-plane heterostructure into the PCSEL.

Indoor occupancy estimation is a critical analytical task for several applications (e.g., social isolation of elderlies). The proliferation of Internet of Things (IoT) devices enabled the occupancy estimation, as it provided access to a mass amount of data. Several works have been proposed exploiting the IoT Passive Inference (PIR) or environmental (e.g., CO2) features. These works however are traditionally selecting the feature space at the learning phase and passively using it over time. Hence, they ignore the dynamics of indoor occupancy, such as the location of the occupant or his motion patterns, leading to a decreasing accuracy over time. In this paper, we study those dynamics and show that motion patterns, along with environmental features favor the occupancy estimation. We design a Location-Aware Hidden Markov Model (HMM), which dynamically adapts the feature space based on the occupant's location. Our experiments on real data show that Location-Aware HMM can reach up to 10% better accuracy than Conventional HMM.

NFV, being discussed and standardized in ETSI, is regarded as a promising candidate technology for Future Networks. Although many advantages are expected, the introduction of NFV will greatly challenge current management systems. In this paper, we summarize some of the carrier's requirements for the management of NFV, based on which we propose vConductor as an NFV management solution. We say vConductor is innovative due to its characteristics of automatic service provisioning, end-to-end inventory management, and generic system design. To achieve these goals, we design a completely modular-based system structure and refine the process of provisioning a virtual network service. Furthermore, we propose a general data model, specifically for NFV, starting from the TM Forum SID. We believe that vConductor will be a catalyst for accelerating the industrialization of NFV.

Capital expenditure (CAPEX) and operating expenses (OPEX) have been practical concerns for telecommunication companies, and Network Function Virtualization (NFV) has attracted significant attention in recent years. However, the NFV Infrastructure (NFVI) includes a wide variety of resource types and stakeholders. Hence, there is still a need for the existing resource scheduling approaches, but they are no longer sufficient. In this paper, we propose a Multi-objective Resource Scheduling Algorithm (MORSA) to optimize the NFVI resources. In the development of the MORSA, we designed a plug-in architecture to satisfy various requirements related to NFVI resources and stakeholder policies. The MORSA allows the NFVI Resource Scheduler to optimize simultaneously the combination of possibly conflicting objectives with multifaceted constraints in complex real world situations. Evaluation results show that the MORSA obtains approximate solutions among conflicting objectives in a reasonable computation time.

Although BitTorrent scales well to support large peer populations, the non-uniform seeder distribution over swarms leads to suboptimal content distribution. In this paper, we propose SeederTrading enabling the over-seeded swarms to trade the seeder bandwidth with the under-seeded swarms to examine the potential improvement through trading seeder bandwidth across swarms. Simulation results show that the under-seeded swarms can improve the content download performance while retaining the performance of over-seeded swarms, which means the content distribution performance and efficiency are improved. We also show that the trading seeder bandwidth across swarms can be done within short time (i.e., 60 seconds in our simulation).

In this paper, we propose a new kind of P2P traffic control technique, called Netpherd exploiting the peer selection adaptation (i.e., preferring peers who are likely to provide better performance). Netpherd tries to enable the peers to communicate with the peers of the local domain by manipulating network performance (i.e., adding an artificial delay to the inter-domain traffic) at network device like router. Simulation results show that Netpherd can increases (decreases) the intra-domain (inter-domain) traffic by affecting the peer selection adaptation. Netpherd also improves the content download performance.

Despite its great success, BitTorrent suffers from a content unavailability problem where peers can not complete their content downloads due to some missing chunks, which is caused by an absence of seeders. Multi-swarm collaboration approach is a natural choice for improving the content avail- ability, since the content unavailability can not be managed by one swarm easily. Most existing multi-swarm collaboration approaches, however, show content-related limitations, which limit their application scopes. In this paper, we introduce a new kind of multi-swarm collaboration utilizing a swarm as a temporal storage. In a nutshell, the collaborating swarms cache some chunks of each other that are likely to be unavailable when the seeders are online and share the cached chunks when the content unavailability happens. Our approach enables any swarms to collaborate with each other without the content-related limitations. Simulation results show that our approach improves the number of download completions by over 50% compared to vanilla BitTorrent with low caching overhead. The results also show that our approach enables the peers participating in our approach to enjoy better performance than other peers, which can be an peer incentive.

Bit Torrent is one of the most popular P2P file sharing applications in the world. Each Bit Torrent network is called a swarm and millions of peers may join multiple swarms. Due to swarm's large network size and complexity, many resources (PC servers, the Internet connection, etc.) are required for measuring all the swarms in the world. For this reason, the existing work is forced to measure only a part of the entire set of swarms, thus, ends up understanding only a part of it. In this paper, we propose a resource-efficient method for crawling multiple Bit Torrent swarms by only a limited amount of resources such as a single PC server. In the proposed method, our crawler avoids collecting redundant information of swarms without pressing WAN access links nor expending much processing resources. We also use a number of techniques to efficiently crawl all the participating peers of multiple swarms. We crawl over 4.3 million unique .torrent files, small files that store metadata used in Bit Torrent, and 48,000 tracker addresses. We can crawl 4.3 million swarms within an hour. We obtain 24 swarm snapshots and 10 million unique peers in a day.

BitTorrent is one of the most popular P2P file sharing applications in the world. Each BitTorrent network is called a swarm and millions of peers may join multiple swarms. However, there are many unreachable peers (NATed, Fire-Walled, or inactive at the time of the measurement) in each swarm. Due to this unreachable peers problem, the existing work can measure only a part of the entire peers in a swarm. In this paper, we propose an improved measurement method for BitTorrent swarms that many unreachable peers. In a nutshell, our crawler obtains peers behind NAT and firewalls by letting them connect to our crawlers through actively advertising our crawlers addresses to them. The evaluation result shows that our proposed method increases the number of unique contacted peers by 112 % compared to the conventional method. The proposed method also increases the total volume of downloaded pieces by 66 %. We then investigate the sampling bias among our proposed method and conventional methods, and find that different measurement methods can lead to significantly different measurement results.

Peer-to-Peer (P2P) file sharing applications have dramatically increased in popularity for the past few years. Although a P2P file sharing network shares many files, it does not usually have a management and control mechanism for the files exchanged. Consequently, copyright infringement and malware propagation in P2P file sharing networks have become prevalent. In order to prevent these file distribution, content poisoning has attracted much attention recently. Although content poisoning aims to obfuscate users by diffusing a lot of poisoned chunks in P2P networks, its effect to the networks have not been well studied yet. In this paper, we apply content poisoning to “Share”, one of the most popular P2P file sharing applications in Japan, to control its file distribution. Our evaluation includes how effective our proposed content poisoning method to a live Share network composed of over 100,000 active peers. The evaluation result shows that our content poisoning method decreases the number of peers that complete file download to less than 5% compared to the case without our control. Our content poisoning method also reduces additional traffic required for poisoning to 4%.

We have proposed a Hamming distance search CAM (content addressable memory) using neuron MOS transistors. The proposed CAM can compare stored data and a search data in parallel, and find stored data with Hamming distance within a certain range. These functions of the CAM are useful for data management and pattern recognition, etc. In this paper, we analyze the power consumption of the proposed CAM in interrogation operation. The proposed CAM can reduce search current to perform the interrogation operation with lower power consumption than conventional NOR-type CAM's with Hamming distance search functions. Furthermore, the results obtained from these analyses are fully confirmed by HSPICE

When a musical session is performed via networks, it is necessary to interact in real-time. There exists, however, the problem of delay between the nodes, which causes musical notes become an impediment. To overcome this problem, we had proposed a new protocol for musical session called mutual anticipated session (M.A.S.), which is a type of ensemble that controls the appropriate timing of the sounds. In the M.A.S, one player's performance precedes the other players', so we call this performance "precedent musical performance", and we call the time lapse between the players' performance as "precedent time." In the current M.A.S, it is assumed that the tempo during the performance is constant. In a case, however, where players want to perform more expressively or more emotionally, they may change the tempo. Thus, in this paper, we add a function that accommodates the change of the tempo and evaluate the function experimentally.

Conventional face image recognition has been done based on the extraction techniques of many features such as shapes and relations among the positions of eyes, mouth, nose and so on. However, it is difficult to extract many features because the human face has an unclear shadow. In addition, the extracted features resemble each other among face images. Therefore, face images recognition for many persons is difficult. In this research, we try to recognize face images for many persons by using a multi-layer neural network and 2-dimensional discrete Walsh transform. This face image recognition is also able to discriminate unlearnt face images.

To represent both the basin-scale circulation in the ocean and the local small-scale variations at the same time, a high resolution global ocean general circulation model (OGCM) is essential. Recent progress in parallel computing techniques together with the high ability of computer itself makes it possible to use such a high resolution OGCM for climate variability issues. We have developed and are using a global OGCM based on the Princeton Ocean Model, with 1/6 degrees horizontal grid resolutions. This model has been running on the Numerical Wind Tunnel at National Aerospace Laboratory in Japan, using 64 PEs with 256 MB memory for each PE, and is successful in reproducing very realistic variations of the currents and tracer fields. In the present study, details of the model and some examples of those variations from 16-year calculation with climatological forcing fields are introduced.

A portable QCD simulation program running on NWT with 128 PE's shows the performance 0.032 micro sec/link update, which is about 189 times faster than a highly optimized code on CRAY X-MP/48 with four processors. The performance corresponds to 178 GFLOPS sustained speed.

NAL Numerical Wind Tunnel (NWT) is a distributed memory parallel computer developed through joint research and development of NAL and Fujitsu. It is based on the analysis of CFD codes developed in NAL. The target performance is more than 100 times faster than VP400. In this paper, the parallel computation model employed in the development of the NWT is described. The specification and feature of the NWT and the NWT Fortran are discussed. Finally, performance evaluations and some applications are presented. We find that the target performance is attained.>