Recent research has demonstrated that Large Language Models (LLMs) can enhance their capabilities by utilizing external tools. However, three pivotal questions remain unanswered: (1) How effective are current LLMs in tools? (2) we LLMs' ability to utilize (3) What obstacles need be overcome leverage To address these questions, introduce API-Bank, a groundbreaking benchmark, specifically designed for tool-augmented LLMs. For the first question, develop runnable evaluation system consisting of...

As Video Streaming and Analytics (VSA) systems become increasingly popular, serious privacy concerns have risen on exposing too much unnecessary private information to the VSA providers. Yet, it is challenging protect while still preserving desired features, i.e., effective analytics, forensic support, resource efficiency, real-time execution. In this paper, we present a enhancement system (PECAM), which addresses above challenge with no change in back-end. PECAM leverages novel Generative...

Background: Sparse angular projection is an important way to reduce CT dose. It consists of two processes, sparse sampling, and image reconstruction based on projection. Under the traditional framework, sparseness angle may cause a degradation effect in reconstructed image. A series machine learning methods for developed recent years, especially deep methods, can effectively improve quality, however, these only reconstruct images certain sampling scheme. Objective: On other words, they...

As the computation bottleneck in lattice-based cryptography (LBC), polynomial multiplication based on number theoretic transform (NTT) has been continuously studied for flexible hardware implementations with high area-efficiency. This paper presents an area-efficient and configurable NTT-based multiplier (AC-PM) incorporating algorithmic architectural level optimization techniques. For core operation of multiplication, two low-complexity fast modular algorithms are introduced loose...

Large language models (LLMs) often contain misleading content, emphasizing the need to align them with human values ensure secure AI systems. Reinforcement learning from feedback (RLHF) has been employed achieve this alignment. However, it encompasses two main drawbacks: (1) RLHF exhibits complexity, instability, and sensitivity hyperparameters in contrast SFT. (2) Despite massive trial-and-error, multiple sampling is reduced pair-wise contrast, thus lacking contrasts a macro perspective. In...

Recently, the National Institute of Standards and Technology (NIST) has identified first four quantum-resistant algorithms for post-quantum cryptography (PQC) standardization. CRYSTALS-Kyber (Kyber) is only public-key encryption key-establishment algorithm among them. In this article, we propose a reconfigurable, high-speed, area-efficient polynomial multiplication accelerator Kyber to facilitate its practical applications. The cornerstone butterfly unit (BU) structure, composed modular...

Recent research has demonstrated that Large Language Models (LLMs) can enhance their capabilities by utilizing external tools. However, three pivotal questions remain unanswered: (1) How effective are current LLMs in tools? (2) we LLMs' ability to utilize (3) What obstacles need be overcome leverage To address these questions, introduce API-Bank, a groundbreaking benchmark, specifically designed for tool-augmented LLMs. For the first question, develop runnable evaluation system consisting of...

Number theoretic transform (NTT) is widely applied as a fundamental component of next-generation cryptosystems. This brief introduces novel high-low interactive memory access pattern for an out-of-place NTT design, which can be configurable in degree parallelism. To achieve high area efficiency, the proposed flexibly selected referring to design parameters. Then first time, we present quantitative analysis about relevance between parallelism and number computing cycles. More importantly, put...

Welding robots application can greatly enhance the production efficiency. And effective welding robot path planning is important for improving efficiency of robot. Due to its simplicity, high search accuracy and fast convergence rate, particle swarm optimization (PSO) algorithm was used study in this paper. A novel hybrid discrete PSO (HDPSO) proposed improve basic after light ray optimization, genetic chaos strategies were combined first. Then, HDPSO solve 3-dimensional problem. Besides, a...

The data abuse issue has risen along with the widespread development of deep learning inference service (DLIS). Specifically, mobile users worry about their input being labeled to secretly train new models that are unrelated DLIS they subscribe to. This unique issue, unlike privacy problem, is rights owners in context learning. However, preventing demanding when considering usability and generality scenario. In this work, we propose, our best knowledge, first prevention mechanism called...

Recently, secure neural network (NN) inference, a combination of homomorphic encryption (HE) and NN, has attracted much attention. Nevertheless, large number computations, mainly brought by the HE scheme, form bottleneck in real-time applications. In this article, we present hardware accelerator on field-programmable gate array (FPGA) for convolution layer (HomConvL), which is most computation-intensive part HE-based inference. First, propose new HomConvL algorithm called packed rotations at...

The lattice-based cryptography (LBC) has been widely used recently in many compute-intensive applications, such as the post-quantum (PQC) and privacy-preserving deep learning, where main task for applications is to improve computational efficiency. modular multiplication operations, mainly involved number theoretic transform (NTT), comprise a large proportion of whole computations required by an LBC. This paper presents novel "decompose-and-reduce" algorithm (DARM), considering primes with...

The Smith-Waterman (S-W) algorithm is widely adopted by the state-of-the-art DNA sequence aligners in next-generation sequencing (NGS). Prevailing read aligners, such as BWA-MEM and Bowtie 2, use S-W to implement seed-and-extend paradigm. In this work, we further extend functionality of evaluate similarity between a pair sequences without going through traceback process, design three-level hardware scoring system compute final result efficiently. made reconfigurable align pairs various...

The increased adoption of electronic health record systems has led to boons for downstream analyses within the clinical domain. identification relevant patient data focal remains a key challenge. In this study, we applied sequential model-based global optimization towards tuning hyperparameters latent Dirichlet allocation, standard topic modeling technique. We showcase physician notes specific chronic lymphocytic leukemia treatment as generalizable use-case. Using each identified component...

The verifiable delay function (VDF), as a kind of cryptographic primitives, has recently been adopted quite often in decentralized systems. Highly correlated to the security VDFs, fastest implementation for VDF evaluation is generally desired be publicly known. In this paper, first time, we propose low-latency hardware complete class group by jointly exploiting optimizations. On one side, reduce required computational cycles decreasing hardware-unfriendly divisions and increase parallelism...

Deep neural networks (DNNs) are the de facto standard for essential use cases, such as image classification, computer vision, and natural language processing. As DNNs datasets get larger, they require distributed training on increasingly larger clusters. A main bottleneck is resulting communication overhead where workers exchange model updates (i.e., gradients) a per-round basis. To address this accelerate training, widely-deployed approach compression. However, previous deployments often...

We present a dissipative scheme to generate an entangled steady-state between two superconducting transmon qutrits separately embedded in coupled transmission line resonators circuit quantum electrodynamics (QED) setup. In our scheme, the resonant qutrit-resonator interaction and photon hopping jointly induce asymmetric energy gaps dressed state subspaces. The coherent driving fields specific transition processes lead gradual accumulation population of target state, combination both drives...

A generating method of long-time-scale photovoltaic (PV) output vector facing power system planning was proposed in the paper. The major feature is that only several kinds accessible data are needed, while four main functions could be realized. In created PV vector, cyclical changes caused by sunrise and sunset considered, overall trend change related to monthly or seasonal differences magnitude leaded weather changing which usually last a few days adjustment make reflecting utilization...