arXiv subjects

Yuhang Yao

Publications and source records attributed to Yuhang Yao.

53 records · Page 3Linked to original sources

arXiv2024

TorchOpera: A Compound AI System for LLM Safety

We introduce TorchOpera, a compound AI system for enhancing the safety and quality of prompts and responses for Large Language Models. TorchOpera ensures that all user prompts are safe, contextually grounded, and effectively processed, while enhancing LLM responses to be relevant and high quality. TorchOpera utilizes the vector database for contextual grounding, rule-based wrappers for flexible modifications, and specialized mechanisms for detecting and adjusting unsafe or incorrect content. We also provide a view of the compound AI system to reduce the computational cost. Extensive experiments show that TorchOpera ensures the safety, reliability, and applicability of LLMs in real-world settings while maintaining the efficiency of LLM responses.

cs.AI↗

arXiv2024

Unlocking high hole mobility in diamond over a wide temperature range via efficient shear strain

As a wide bandgap semiconductor, diamond holds both excellent electrical and thermal properties, making it highly promising in the electrical industry. However, its hole mobility is relatively low and dramatically decreases with increasing temperature, which severely limits further applications. Herein, we proposed that the hole mobility can be efficiently enhanced via slight compressive shear strain along the [100] direction, while the improvement via shear strain along the [111] direction is marginal. This impressive distinction is attributed to the deformation potential and the elastic compliance matrix. The shear strain breaks the symmetry of the crystalline structure and lifts the band degeneracy near the valence band edge, resulting in a significant suppression of interband electron-phonon scattering. Moreover, the hole mobility becomes less temperature-dependent due to the decrease of electron scatterings from high-frequency acoustic phonons. Remarkably, the in-plane hole mobility of diamond is increased by approximately 800% at 800 K with a 2% compressive shear strain along the [100] direction. The efficient shear strain strategy can be further extended to other semiconductors with face-centered cubic geometry.

cond-mat.mtrl-sci↗

arXiv2024

TensorOpera Router: A Multi-Model Router for Efficient LLM Inference

With the rapid growth of Large Language Models (LLMs) across various domains, numerous new LLMs have emerged, each possessing domain-specific expertise. This proliferation has highlighted the need for quick, high-quality, and cost-effective LLM query response methods. Yet, no single LLM exists to efficiently balance this trilemma. Some models are powerful but extremely costly, while others are fast and inexpensive but qualitatively inferior. To address this challenge, we present TO-Router, a non-monolithic LLM querying system that seamlessly integrates various LLM experts into a single query interface and dynamically routes incoming queries to the most high-performant expert based on query's requirements. Through extensive experiments, we demonstrate that when compared to standalone expert models, TO-Router improves query efficiency by up to 40\%, and leads to significant cost reductions of up to 30%, while maintaining or enhancing model performance by up to 10%.

cs.AI↗

arXiv2024

Superoscillation focusing of high-order cylindrical-vector beams

Traditional superoscillation focusing typically requires complex optimization of the incident light field. These complexities may limit the practical application of superoscillation. High-order radially polarized Laguerre-Gaussian beams inherently support superoscillation focusing due to their multi-ring amplitude distribution and 0 ~ πphase alternation, which align with the necessary destructive interference mechanisms. In this study, we demonstrate that by adjusting the beam mode order together with the incident beam size, we can easily control the full width at half maximum, field of view, and energy distribution of superoscillation focusing. Moreover, high-order azimuthally polarized vortex-phase Laguerre-Gaussian beams can also achieve superoscillation focusing, offering even better super-resolution effects. The distinct focusing behaviors of their circular components present unique opportunities for applications involving circular dichroism materials.

physics.optics↗

arXiv2024

ScaleLLM: A Resource-Frugal LLM Serving Framework by Optimizing End-to-End Efficiency

Large language models (LLMs) have surged in popularity and are extensively used in commercial applications, where the efficiency of model serving is crucial for the user experience. Most current research focuses on optimizing individual sub-procedures, e.g. local inference and communication, however, there is no comprehensive framework that provides a holistic system view for optimizing LLM serving in an end-to-end manner. In this work, we conduct a detailed analysis to identify major bottlenecks that impact end-to-end latency in LLM serving systems. Our analysis reveals that a comprehensive LLM serving endpoint must address a series of efficiency bottlenecks that extend beyond LLM inference. We then propose ScaleLLM, an optimized system for resource-efficient LLM serving. Our extensive experiments reveal that with 64 concurrent requests, ScaleLLM achieves a 4.3x speed up over vLLM and outperforms state-of-the-arts with 1.5x higher throughput.

cs.DC↗

arXiv2024

FedSecurity: Benchmarking Attacks and Defenses in Federated Learning and Federated LLMs

This paper introduces FedSecurity, an end-to-end benchmark that serves as a supplementary component of the FedML library for simulating adversarial attacks and corresponding defense mechanisms in Federated Learning (FL). FedSecurity eliminates the need for implementing the fundamental FL procedures, e.g., FL training and data loading, from scratch, thus enables users to focus on developing their own attack and defense strategies. It contains two key components, including FedAttacker that conducts a variety of attacks during FL training, and FedDefender that implements defensive mechanisms to counteract these attacks. FedSecurity has the following features: i) It offers extensive customization options to accommodate a broad range of machine learning models (e.g., Logistic Regression, ResNet, and GAN) and FL optimizers (e.g., FedAVG, FedOPT, and FedNOVA); ii) it enables exploring the effectiveness of attacks and defenses across different datasets and models; and iii) it supports flexible configuration and customization through a configuration file and some APIs. We further demonstrate FedSecurity's utility and adaptability through federated training of Large Language Models (LLMs) to showcase its potential on a wide range of complex applications.

cs.CR↗

arXiv2024

FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System

Federated Learning trains machine learning models on distributed devices by aggregating local model updates instead of local data. However, privacy concerns arise as the aggregated local models on the server may reveal sensitive personal information by inversion attacks. Privacy-preserving methods, such as homomorphic encryption (HE), then become necessary for FL training. Despite HE's privacy advantages, its applications suffer from impractical overheads, especially for foundation models. In this paper, we present FedML-HE, the first practical federated learning system with efficient HE-based secure model aggregation. FedML-HE proposes to selectively encrypt sensitive parameters, significantly reducing both computation and communication overheads during training while providing customizable privacy preservation. Our optimized system demonstrates considerable overhead reduction, particularly for large foundation models (e.g., ~10x reduction for ResNet-50, and up to ~40x reduction for BERT), demonstrating the potential for scalable HE-based FL deployment.

cs.LG↗

arXiv2024

The Capacity of 3 User Linear Computation Broadcast

The $K$ User Linear Computation Broadcast (LCBC) problem is comprised of $d$ dimensional data (from $\mathbb{F}_q$), that is fully available to a central server, and $K$ users, who require various linear computations of the data, and have prior knowledge of various linear functions of the data as side-information. The optimal broadcast cost is the minimum number of $q$-ary symbols to be broadcast by the server per computation instance, for every user to retrieve its desired computation. The reciprocal of the optimal broadcast cost is called the capacity. The main contribution of this paper is the exact capacity characterization for the $K=3$ user LCBC for all cases, i.e., for arbitrary finite fields $\mathbb{F}_q$, arbitrary data dimension $d$, and arbitrary linear side-informations and demands at each user. A remarkable aspect of the converse is that unlike the $2$ user LCBC whose capacity was determined previously, the entropic formulation (where the entropies of demands and side-informations are specified, but not their functional forms) is insufficient to obtain a tight converse for the $3$ user LCBC. Instead, the converse exploits functional submodularity. Notable aspects of achievability include a decomposition of the users' collective signal space into subspaces that allow different degrees of efficiency in broadcast cost, revealing a tradeoff that leads to a constrained water-filling solution. Random coding arguments are invoked to resolve compatibility issues that arise as each user has a different view of these subspaces, conditioned on its own side-information.

cs.IT↗

arXiv2024

The Capacity of Classical Summation over a Quantum MAC with Arbitrarily Distributed Inputs and Entanglements

The $Σ$-QMAC problem is introduced, involving $S$ servers, $K$ classical ($\mathbb{F}_d$) data streams, and $T$ independent quantum systems. Data stream ${\sf W}_k, k\in[K]$ is replicated at a subset of servers $\mathcal{W}(k)\subset[S]$, and quantum system $\mathcal{Q}_t, t\in[T]$ is distributed among a subset of servers $\mathcal{E}(t)\subset[S]$ such that Server $s\in\mathcal{E}(t)$ receives subsystem $\mathcal{Q}_{t,s}$ of $\mathcal{Q}_t=(\mathcal{Q}_{t,s})_{s\in\mathcal{E}(t)}$. Servers manipulate their quantum subsystems according to their data and send the subsystems to a receiver. The total download cost is $\sum_{t\in[T]}\sum_{s\in\mathcal{E}(t)}\log_d|\mathcal{Q}_{t,s}|$ qudits, where $|\mathcal{Q}|$ is the dimension of $\mathcal{Q}$. The states and measurements of $(\mathcal{Q}_t)_{t\in[T]}$ are required to be separable across $t\in[T]$ throughout, but for each $t\in[T]$, the subsystems of $\mathcal{Q}_{t}$ can be prepared initially in an arbitrary (independent of data) entangled state, manipulated arbitrarily by the respective servers, and measured jointly by the receiver. From the measurements, the receiver must recover the sum of all data streams. Rate is defined as the number of dits ($\mathbb{F}_d$ symbols) of the desired sum computed per qudit of download. The capacity of $Σ$-QMAC, i.e., the supremum of achievable rates is characterized for arbitrary data replication and entanglement distribution maps $\mathcal{W}, \mathcal{E}$. Coding based on the $N$-sum box abstraction is optimal in every case. Notably, for every $S\neq 3$ there exists an instance of the $Σ$-QMAC where $S$-party entanglement is necessary to achieve the fully entangled capacity.

cs.IT↗

arXiv2024

Revisiting phonon thermal transport in two-dimensional gallium nitride: higher-order phonon-phonon and phonon-electron scattering

Two-dimensional gallium nitride (2D-GaN) has great potential in power electronics and optoelectronics. Heat dissipation is a critical issue for these applications of 2D-GaN. Previous studies showed that higher-order phonon-phonon scattering has extremely strong effects on the lattice thermal conductivity of 2D-GaN, which exhibits noticeable discrepancies with lattice thermal conductivity calculated from molecular dynamics. In this work, it is found that the fourth-order interatomic force constants (4th-IFCs) of 2D-GaN are quite sensitive to atomic displacement in the finite different method. The effects of the four-phonon scattering can be severely overestimated with non-convergent 4th-IFCs. The lattice thermal conductivity from three-phonon scattering is reduced by 65.6% due to four-phonon scattering. The reflection symmetry allows significantly more four-phonon processes than three-phonon processes. It was previously thought the electron-phonon interactions have significant effects on the lattice thermal conductivity of two-dimensional materials. However, the effects of phonon-electron interactions on the lattice thermal conductivity of both n-type and p-type 2D-GaN at high charge carrier concentrations can be neglected due to the few phonon-electron scattering channels and the relatively strong four-phonon scattering.

cond-mat.mtrl-sci↗

arXiv2023

FedGCN: Convergence-Communication Tradeoffs in Federated Training of Graph Convolutional Networks

Methods for training models on graphs distributed across multiple clients have recently grown in popularity, due to the size of these graphs as well as regulations on keeping data where it is generated. However, the cross-client edges naturally exist among clients. Thus, distributed methods for training a model on a single graph incur either significant communication overhead between clients or a loss of available information to the training. We introduce the Federated Graph Convolutional Network (FedGCN) algorithm, which uses federated learning to train GCN models for semi-supervised node classification with fast convergence and little communication. Compared to prior methods that require extra communication among clients at each training round, FedGCN clients only communicate with the central server in one pre-training step, greatly reducing communication costs and allowing the use of homomorphic encryption to further enhance privacy. We theoretically analyze the tradeoff between FedGCN's convergence rate and communication cost under different data distributions. Experimental results show that our FedGCN algorithm achieves better model accuracy with 51.7% faster convergence on average and at least 100X less communication compared to prior work.

cs.LG↗

arXiv2023

$N$-Sum Box: An Abstraction for Linear Computation over Many-to-one Quantum Networks

Linear computations over quantum many-to-one communication networks offer opportunities for communication cost improvements through schemes that exploit quantum entanglement among transmitters to achieve superdense coding gains, combined with classical techniques such as interference alignment. The problem becomes much more broadly accessible if suitable abstractions can be found for the underlying quantum functionality via classical black box models. This work formalizes such an abstraction in the form of an "$N$-sum box", a black box generalization of a two-sum protocol of Song \emph{et al.} with recent applications to $N$-server private information retrieval. The $N$-sum box has a communication cost of $N$ qudits and classical output of a vector of $N$ $q$-ary digits linearly dependent (via an $N \times 2N$ transfer matrix) on $2N$ classical inputs distributed among $N$ transmitters. We characterize which transfer matrices are feasible by our construction, both with and without the possibility of additional locally invertible classical operations at the transmitters and receivers. Furthermore, we provide a sample application to Cross-Subspace Alignment (CSA) schemes to obtain efficient instances of Quantum Private Information Retrieval (QPIR) and Quantum Secure Distributed Batch Matrix Multiplication (QSDBMM). We first describe $N$-sum boxes based on maximal stabilizers and we then consider non-maximal-stabilizer-based constructions to obtain an instance of Quantum Symmetric Private Information Retrieval.

cs.IT↗

arXiv2022

FedRule: Federated Rule Recommendation System with Graph Neural Networks

Much of the value that IoT (Internet-of-Things) devices bring to ``smart'' homes lies in their ability to automatically trigger other devices' actions: for example, a smart camera triggering a smart lock to unlock a door. Manually setting up these rules for smart devices or applications, however, is time-consuming and inefficient. Rule recommendation systems can automatically suggest rules for users by learning which rules are popular based on those previously deployed (e.g., in others' smart homes). Conventional recommendation formulations require a central server to record the rules used in many users' homes, which compromises their privacy and leaves them vulnerable to attacks on the central server's database of rules. Moreover, these solutions typically leverage generic user-item matrix methods that do not fully exploit the structure of the rule recommendation problem. In this paper, we propose a new rule recommendation system, dubbed as FedRule, to address these challenges. One graph is constructed per user upon the rules s/he is using, and the rule recommendation is formulated as a link prediction task in these graphs. This formulation enables us to design a federated training algorithm that is able to keep users' data private. Extensive experiments corroborate our claims by demonstrating that FedRule has comparable performance as the centralized setting and outperforms conventional solutions.

cs.LG↗

arXiv2022

Faithful Explanations for Deep Graph Models

This paper studies faithful explanations for Graph Neural Networks (GNNs). First, we provide a new and general method for formally characterizing the faithfulness of explanations for GNNs. It applies to existing explanation methods, including feature attributions and subgraph explanations. Second, our analytical and empirical results demonstrate that feature attribution methods cannot capture the nonlinear effect of edge features, while existing subgraph explanation methods are not faithful. Third, we introduce \emph{k-hop Explanation with a Convolutional Core} (KEC), a new explanation method that provably maximizes faithfulness to the original GNN by leveraging information about the graph structure in its adjacency matrix and its \emph{k-th} power. Lastly, our empirical results over both synthetic and real-world datasets for classification and anomaly detection tasks with GNNs demonstrate the effectiveness of our approach.

cs.LG↗

arXiv2021

Electromagnetic Proton Beam Instabilities in the Inner Heliosphere: Energy Transfer Rate, Radial Distribution, and Effective Excitation

Differential flows among different ion species are often observed in the solar wind, and such ion differential flows can provide the free energy to drive Alfvén/ion-cyclotron and fast-magnetosonic/whistler instabilities. Previous works mainly focused on ion beam instability under the parameters representative of the solar wind nearby 1 au. In this paper we further study proton beam instability using the radial models of the magnetic field and plasma parameters in the inner heliosphere. We explore a comprehensive distribution of proton beam instability as functions of the heliocentric distance and the beam speed. We also perform a detailed analysis of the energy transfer between unstable waves and particles and quantify how much the free energy of the proton beam flows into unstable waves and other kinds of particle species (i.e., proton core, alpha particle, and electron). This work clarifies that both parallel and perpendicular electric fields are responsible for the excitation of oblique Alfvén/ion-cyclotron and oblique fast-magnetosonic/whistler instabilities. Moreover, this work proposes an effective growth length to estimate whether the instability is efficiently excited or not. It shows that oblique Alfvén/ion-cyclotron instability, oblique fast-magnetosonic/whistler instability, and oblique Alfvén/ion-beam instability can be efficiently driven by proton beams drifting at the speed $\sim$ 600-1300 km s$^{-1}$ in the solar atmosphere. In particular, oblique Alfvén/ion-cyclotron waves driven in the solar atmosphere can be significantly damped therein, leading to the solar corona heating. These results are helpful for understanding proton beam dynamics in the inner heliosphere and can be verified through in situ satellite measurements.

astro-ph.SR↗

arXiv2021

GCN-SE: Attention as Explainability for Node Classification in Dynamic Graphs

Graph Convolutional Networks (GCNs) are a popular method from graph representation learning that have proved effective for tasks like node classification tasks. Although typical GCN models focus on classifying nodes within a static graph, several recent variants propose node classification in dynamic graphs whose topologies and node attributes change over time, e.g., social networks with dynamic relationships, or literature citation networks with changing co-authorships. These works, however, do not fully address the challenge of flexibly assigning different importance to snapshots of the graph at different times, which depending on the graph dynamics may have more or less predictive power on the labels. We address this challenge by proposing a new method, GCN-SE, that attaches a set of learnable attention weights to graph snapshots at different times, inspired by Squeeze and Excitation Net (SE-Net). We show that GCN-SE outperforms previously proposed node classification methods on a variety of graph datasets. To verify the effectiveness of the attention weight in determining the importance of different graph snapshots, we adapt perturbation-based methods from the field of explainable machine learning to graphical settings and evaluate the correlation between the attention weights learned by GCN-SE and the importance of different snapshots over time. These experiments demonstrate that GCN-SE can in fact identify different snapshots' predictive power for dynamic node classification.

cs.LG↗

arXiv2021

Interpretable Clustering on Dynamic Graphs with Recurrent Graph Neural Networks

We study the problem of clustering nodes in a dynamic graph, where the connections between nodes and nodes' cluster memberships may change over time, e.g., due to community migration. We first propose a dynamic stochastic block model that captures these changes, and a simple decay-based clustering algorithm that clusters nodes based on weighted connections between them, where the weight decreases at a fixed rate over time. This decay rate can then be interpreted as signifying the importance of including historical connection information in the clustering. However, the optimal decay rate may differ for clusters with different rates of turnover. We characterize the optimal decay rate for each cluster and propose a clustering method that achieves almost exact recovery of the true clusters. We then demonstrate the efficacy of our clustering algorithm with optimized decay rates on simulated graph data. Recurrent neural networks (RNNs), a popular algorithm for sequence learning, use a similar decay-based method, and we use this insight to propose two new RNN-GCN (graph convolutional network) architectures for semi-supervised graph clustering. We finally demonstrate that the proposed architectures perform well on real data compared to state-of-the-art graph clustering algorithms.

cs.LG↗