arXiv subjects

Yitao Chen

Publications and source records attributed to Yitao Chen.

At least 19 recordsLinked to original sources

arXiv2026

How Do Prompt Variations Affect Energy Consumption in On-Device LLMs?

Large language models (LLMs) are increasingly deployed on mobile devices, making energy efficiency a key deployment constraint, yet the energy impact of prompt design remains underexplored. This paper aims to understand how two prompt properties, cognitive load and phrasing pattern, shape the energy behavior of on-device LLM inference. We conduct a broad empirical study covering prompt properties, datasets, models, and devices, with phase-level profiling that separates prefill and decode energy. We find that cognitive load primarily affects the energy cost per token, while phrasing pattern affects energy largely through token usage. Our energy-quality analysis further shows that prompt design reshapes the attainable frontier differently across models, highlighting the need for model-aware prompt design in energy-efficient on-device LLM inference. Code, datasets, and scripts are available at https://amai-gsu.github.io/PromptProperty/.

cs.CL↗

arXiv2026

Real-time decoding of quantum error correction codes using high-performance computing

Quantum error correction (QEC) is indispensable for building scalable fault-tolerant quantum computers. Effective QEC demands stringent real-time decoding: the decoder must process syndrome measurements and determine corrections within a time scale--typically on the order of microseconds, to avoid data backlog. Scaling to large number of logical qubits further necessitates significant computational resources. In this work, we propose an architecture, called \emph{THQLink}, for real-time decoding of quantum error correction codes using high-performance computing (HPC) resources. The network connecting the HPC and the control system of quantum processing unit (QPU) is built on TH-Express and can be adapted to different quantum technologies and their associated control stacks. We report a round-trip latency of 2.944 $μ$s on average, with an incremental overhead of 130 ns per additional hop. Using a parallel window strategy, we demonstrate real-time decoding (1 $μ$s per QEC round) of the surface code up to distance 19 using a matching-based decoder on CPUs. Our work presents a scalable framework for real-time decoding in fault-tolerant quantum computing. It can be readily applied to quantum-centric supercomputers that feature tight integration between QPU and HPC resources, thereby enabling efficient support for hybrid quantum-classical algorithms and computation-intensive workloads offloaded from the QPU.

quant-ph↗

arXiv2026

InitGen: Candidate Generation for Interaction Initiation in Intelligent Assistants

Interaction initiation refers to presenting multiple candidate queries when a user opens an intelligent assistant before expressing any intent for the current session. In production, candidate generation incorporates dynamic context and produces all candidates within a strict latency budget. Learning from user feedback is also difficult since the generator usually produces more candidates than are finally displayed. After downstream filtering and ranking, only a subset is exposed to users, so the observed feedback is partial and cannot be reliably assigned to individual queries. We present InitGen, a framework for candidate generation that is deployed in the interaction initiation pipeline of OPPO's Xiaobu Assistant. InitGen generates a set of candidate queries jointly and aligns the generated set with user feedback through weighted preference optimization. The sample weights are derived from user activity and downstream ranking scores. The activity weight reduces the dominance of highly active users during training, while the ranking score is used as a practical estimate of the reliability of the observed feedback. InitGen also uses a rolling window update strategy to incorporate recent interaction data into periodic model updates. In an online A/B test against a strong production baseline, InitGen improves the click-through rate from 0.95% to 1.61%, corresponding to a relative improvement of 69.1%, and increases query exposure by 17.9% under the same traffic allocation. InitGen generates the complete candidate set within 180 ms and has been fully deployed in OPPO's Xiaobu Assistant, which serves over 150 million monthly active users.

cs.IR↗

arXiv2026

EdgeFaaS: A Function-based Framework for Edge Computing

Edge computing brings unique challenges as the resources on the edge are highly diverse in capabilities and capacities, and highly distributed across many users and the physical world. Existing distributed computing frameworks cannot adequately handle this level of heterogeneity and distribution. This paper proposes EdgeFaaS, a novel function-based edge computing framework to enable edge applications to effectively utilize heterogeneous resources distributed across the Internet of Things (IoT), edge, and cloud for computing. It proposes function virtualization and storage virtualization to abstract distributed and heterogeneous physical resources and provides consistent virtual interfaces for deploying and executing functions and storing and accessing data. EdgeFaaS provides comprehensive support to diverse edge computing workflows, and at the same time allows users to flexibly adjust the configurations and explore various important tradeoffs. To demonstrate its usability, the paper also presents the implementation and evaluation of three representative workflows on EdgeFaaS for video analytics, federated learning, and audio classification, on a real testbed of 100+ geographically distributed IoT devices, edge servers, and cloud services. EdgeFaaS allows users to flexibly explore the deployment configurations of these workflows over distributed and heterogeneous resources. For example, users can easily vary the function placement of the video processing pipeline across IoT, edge, and cloud resources and study the tradeoff between computation and communication costs; users can also flexibly adjust the cluster count and size in the hierarchical federated learning system and explore the tradeoff between training accuracy and speed.

cs.DC↗

arXiv2026

Efficient Diffusion Distillation via Embedding Loss

Recent advances in distilling expensive diffusion models into efficient few-step generators show significant promise. However, these methods typically demand substantial computational resources and extended training periods, limiting accessibility for resource-constrained researchers, and existing supplementary loss functions have notable limitations. Regression loss requires pre-generating large datasets before training and limits the student model to the teacher's performance, while GAN-based losses suffer from training instability and require careful tuning. In this paper, we propose Embedding Loss (EL), a novel supplementary loss function that complements existing diffusion distillation methods to enhance generation quality and accelerate training with smaller batch sizes. Leveraging feature embeddings from a diverse set of randomly initialized networks, EL effectively aligns the feature distributions between the distilled few-step generator and the original data. By computing Maximum Mean Discrepancy (MMD) in the embedded feature space, EL ensures robust distribution matching, thereby preserving sample fidelity and diversity during distillation. Within distribution matching distillation frameworks, EL demonstrates strong empirical performance for one-step generators. On the CIFAR-10 dataset, our approach achieves state-of-the-art FID values of 1.475 for unconditional generation and 1.380 for conditional generation. Beyond CIFAR-10, we further validate EL across multiple benchmarks and distillation methods, including ImageNet, AFHQ-v2, and FFHQ datasets, using DMD, DI, and CM distillation frameworks, demonstrating consistent improvements over existing one-step distillation methods. Our method also reduces training iterations by up to 80%, offering a more practical and scalable solution for deploying diffusion-based generative models in resource-constrained environments.

cs.CV↗

arXiv2025

Scene-Aware Conversational ADAS with Generative AI for Real-Time Driver Assistance

While autonomous driving technologies continue to advance, current Advanced Driver Assistance Systems (ADAS) remain limited in their ability to interpret scene context or engage with drivers through natural language. These systems typically rely on predefined logic and lack support for dialogue-based interaction, making them inflexible in dynamic environments or when adapting to driver intent. This paper presents Scene-Aware Conversational ADAS (SC-ADAS), a modular framework that integrates Generative AI components including large language models, vision-to-text interpretation, and structured function calling to enable real-time, interpretable, and adaptive driver assistance. SC-ADAS supports multi-turn dialogue grounded in visual and sensor context, allowing natural language recommendations and driver-confirmed ADAS control. Implemented in the CARLA simulator with cloud-based Generative AI, the system executes confirmed user intents as structured ADAS commands without requiring model fine-tuning. We evaluate SC-ADAS across scene-aware, conversational, and revisited multi-turn interactions, highlighting trade-offs such as increased latency from vision-based context retrieval and token growth from accumulated dialogue history. These results demonstrate the feasibility of combining conversational reasoning, scene perception, and modular ADAS control to support the next generation of intelligent driver assistance.

cs.RO↗

arXiv2024

Moiré superlattices of antimonene on a Bi(111) substrate with van Hove singularity and Rashba-type spin polarization

Moiré superlattices consisting of two-dimensional materials have attracted immense attention because of emergent phenomena such as flat band-induced Mott insulating states and unconventional superconductivity. However, the effects of spin-orbit coupling on these materials have not yet been fully explored. Here, we show that single- and double-bilayer antimony honeycomb lattices, referred to as antimonene, form moiré superlattices on a Bi(111) substrate due to lattice mismatch. Scanning tunnelling microscopy (STM) measurements reveal the presence of spectral peaks near the Fermi level, which are spatially modulated with the moiré period. Angle-resolved photoemission spectroscopy (ARPES) combined with density functional theory calculations clarify the surface band structure with saddle points near the Fermi level, which allows us to attribute the observed STM spectral peaks to the van Hove singularity. Moreover, spin-resolved ARPES measurements reveal that the observed surface states are Rashba-type spin-polarized. The present work has significant implications in that Fermi surface instability and symmetry breaking may emerge at low temperatures, where the spin degree of freedom and electron correlation also play important roles.

cond-mat.mes-hall↗

arXiv2023

LiDAR-based Outdoor Crowd Management for Smart Campus on the Edge

Crowd management is crucial for a smart campus. Popular methods are camera-based. However, conventional camera-based approaches may leak users' personally identifiable features, jeopardizing user's privacy, which limits its application. In this work, we investigate using affordable light detection and ranging (LiDAR) technology to perform outdoor crowd management leveraging edge computing. Specifically, we aim to count the number of people on a walkway of a university campus. Besides privacy protection, LiDAR sensors are superior to cameras since their performance will not be compromised when the campus is not well-illuminated. We deploy LiDAR sensors on light poles to collect data from the crowd on the campus and leverage edge accelerators to process data locally. We proposed two different methodologies in this work: 1) a non-convolutional neural network (CNN)-based approach, using clustering and autoencoder, and 2) a CNN-based approach that first projects point clouds to 2D planes and then processes the projection with conventional CNNs. Our first approach relies on careful feature engineering, whereas our second approach does not require such effort. However, the CNN-based approach requires more computational power than our non-CNN-based approach. We evaluate both approaches comprehensively with our hand-labeled real-life data collected from campus. Our evaluation results show that the first method achieves an accuracy of 85.4%, whereas the second method achieves 95.8%. Our CNN-based method outperforms existing solutions significantly. We also deploy our two models on an edge accelerator, TPU, to measure the speedup, leveraging this specialized accelerator.

cs.CV↗

arXiv2023

Confidence-based federated distillation for vision-based lane-centering

A fundamental challenge of autonomous driving is maintaining the vehicle in the center of the lane by adjusting the steering angle. Recent advances leverage deep neural networks to predict steering decisions directly from images captured by the car cameras. Machine learning-based steering angle prediction needs to consider the vehicle's limitation in uploading large amounts of potentially private data for model training. Federated learning can address these constraints by enabling multiple vehicles to collaboratively train a global model without sharing their private data, but it is difficult to achieve good accuracy as the data distribution is often non-i.i.d. across the vehicles. This paper presents a new confidence-based federated distillation method to improve the performance of federated learning for steering angle prediction. Specifically, it proposes the novel use of entropy to determine the predictive confidence of each local model, and then selects the most confident local model as the teacher to guide the learning of the global model. A comprehensive evaluation of vision-based lane centering shows that the proposed approach can outperform FedAvg and FedDF by 11.3% and 9%, respectively.

cs.CV↗

arXiv2023

A Contrastive Knowledge Transfer Framework for Model Compression and Transfer Learning

Knowledge Transfer (KT) achieves competitive performance and is widely used for image classification tasks in model compression and transfer learning. Existing KT works transfer the information from a large model ("teacher") to train a small model ("student") by minimizing the difference of their conditionally independent output distributions. However, these works overlook the high-dimension structural knowledge from the intermediate representations of the teacher, which leads to limited effectiveness, and they are motivated by various heuristic intuitions, which makes it difficult to generalize. This paper proposes a novel Contrastive Knowledge Transfer Framework (CKTF), which enables the transfer of sufficient structural knowledge from the teacher to the student by optimizing multiple contrastive objectives across the intermediate representations between them. Also, CKTF provides a generalized agreement to existing KT techniques and increases their performance significantly by deriving them as specific cases of CKTF. The extensive evaluation shows that CKTF consistently outperforms the existing KT works by 0.04% to 11.59% in model compression and by 0.4% to 4.75% in transfer learning on various models and datasets.

cs.LG↗

arXiv2022

Compactifications of moduli space of (quasi-)trielliptic K3 surfaces

We study the moduli space $\mathcal{F}_{T_1}$ of quasi-trielliptic K3 surfaces of type I, whose general member is a smooth bidegree $(2,3)$-hypersurface of $\mathbb{P}^1\times \mathbb{P}^2$. Such moduli space plays an important role in the study of the Hassett-Keel-Looijenga program of the moduli space of degree $8$ quasi-polarized K3 surfaces. In this paper, we consider several natural compactifications of $\mathcal{F}_{T_1}$, such as the GIT compactification and arithmetic compactifications. We give a complete analysis of GIT stability of $(2,3)$-hypersurfaces and provide a concrete description of the boundary of the GIT compactification. For the Baily--Borel compactification of the quasi-trielliptic K3 surfaces, we also compute the configurations of the boundary by classifying certain lattice embeddings. As an application, we show that $(\mathbb{P}^1\times \mathbb{P}^2,εS)$ with small $ε$ is K-stable if $S$ is a K3 surface with at worst ADE singularities. This gives a concrete description of the boundary of the K-stability compactification via the identification of the GIT stability and the K-stability. We also discuss the connection between the GIT, Baily--Borel compactification, and Looijenga's compactifications by studying the projective models of quasi-trielliptic K3 surfaces.

math.AG↗

arXiv2022

Enabling Deep Learning on Edge Devices through Filter Pruning and Knowledge Transfer

Deep learning models have introduced various intelligent applications to edge devices, such as image classification, speech recognition, and augmented reality. There is an increasing need of training such models on the devices in order to deliver personalized, responsive, and private learning. To address this need, this paper presents a new solution for deploying and training state-of-the-art models on the resource-constrained devices. First, the paper proposes a novel filter-pruning-based model compression method to create lightweight trainable models from large models trained in the cloud, without much loss of accuracy. Second, it proposes a novel knowledge transfer method to enable the on-device model to update incrementally in real time or near real time using incremental learning on new data and enable the on-device model to learn the unseen categories with the help of the in-cloud model in an unsupervised fashion. The results show that 1) our model compression method can remove up to 99.36% parameters of WRN-28-10, while preserving a Top-1 accuracy of over 90% on CIFAR-10; 2) our knowledge transfer method enables the compressed models to achieve more than 90% accuracy on CIFAR-10 and retain good accuracy on old categories; 3) it allows the compressed models to converge within real time (three to six minutes) on the edge for incremental learning tasks; 4) it enables the model to classify unseen categories of data (78.92% Top-1 accuracy) that it is never trained with.

cs.LG↗

arXiv2021

Convergence of Generalized Belief Propagation Algorithm on Graphs with Motifs

Belief propagation is a fundamental message-passing algorithm for numerous applications in machine learning. It is known that belief propagation algorithm is exact on tree graphs. However, belief propagation is run on loopy graphs in most applications. So, understanding the behavior of belief propagation on loopy graphs has been a major topic for researchers in different areas. In this paper, we study the convergence behavior of generalized belief propagation algorithm on graphs with motifs (triangles, loops, etc.) We show under a certain initialization, generalized belief propagation converges to the global optimum of the Bethe free energy for ferromagnetic Ising models on graphs with motifs.

cs.LG↗

arXiv2020

Multi-Agent Decentralized Belief Propagation on Graphs

We consider the problem of interactive partially observable Markov decision processes (I-POMDPs), where the agents are located at the nodes of a communication network. Specifically, we assume a certain message type for all messages. Moreover, each agent makes individual decisions based on the interactive belief states, the information observed locally and the messages received from its neighbors over the network. Within this setting, the collective goal of the agents is to maximize the globally averaged return over the network through exchanging information with their neighbors. We propose a decentralized belief propagation algorithm for the problem, and prove the convergence of our algorithm. Finally we show multiple applications of our framework. Our work appears to be the first study of decentralized belief propagation algorithm for networked multi-agent I-POMDPs.

cs.AI↗

arXiv2018

On the Key Generation Rate of Physically Unclonable Functions

In this paper, an algebraic binning based coding scheme and its associated achievable rate for key generation using physically unclonable functions (PUFs) is determined. This achievable rate is shown to be optimal under the generated-secret (GS) model for PUFs. Furthermore, a polar code based polynomial-time encoding and decoding scheme that achieves this rate is also presented.

cs.IT↗

arXiv2018

From Centralized to Decentralized Coded Caching

We consider the problem of designing decentralized schemes for coded caching. In this problem there are $K$ users each caching $M$ files out of a library of $N$ total files. The question is to minimize $R$, the number of broadcast transmissions to satisfy all the user demands. Decentralized schemes allow the creation of each cache independently, allowing users to join or leave without dependencies. Previous work showed that to achieve a coding gain $g$, i.e. $R \leq K (1-M/N)/g$ transmissions, each file has to be divided into number of subpackets that is exponential in $g$. In this work we propose a simple translation scheme that converts any constant rate centralized scheme into a random decentralized placement scheme that guarantees a target coding gain of $g$. If the file size in the original constant rate centralized scheme is subexponential in $K$, then the file size for the resulting scheme is subexponential in $g$. When new users join, the rest of the system remains the same. However, we require an additional communication overhead of $O(\log K)$ bits to determine the new user's cache state. We also show that the worst-case rate guarantee degrades only by a constant factor due to the dynamics of user arrival and departure.

cs.IT↗

arXiv2017

Reconciling Selfish Routing with Social Good

Selfish routing is a central problem in algorithmic game theory, with one of the principal applications being that of routing in road networks. Inspired by the emergence of routing technologies and autonomous driving, we revisit selfish routing and consider three possible outcomes of it: (i) $θ$-Positive Nash Equilibrium flow, where every path that has non-zero flow on all of its edges has cost no greater than $θ$ times the cost of any other path, (ii) $θ$-Used Nash Equilibrium flow, where every used path that appears in the path flow decomposition has cost no greater than $θ$ times the cost of any other path, and (iii) $θ$-Envy Free flow, where every path that appears in the path flow decomposition has cost no greater than $θ$ times the cost of any other path in the path flow decomposition. We first examine the relations of these outcomes among each other and then measure their possible impact on the network's performance. Afterwards, we examine the computational complexity of finding such flows of minimum social cost and give a range for $θ$ for which this task is easy and a range for $θ$ for which this task is NP-hard. Finally, we propose deterministic strategies which, in a worst case approach, can be used by a central planner in order to provide good such flows, and further introduce a natural idea for randomly routing players after giving them specific guarantees about their costs in the randomized routing, as a tool for the central planner to implement a desired flow.

cs.GT↗

arXiv2017

Approximate Capacity of a Class of Partially Connected Interference Channels

We derive inner and outer bounds on the capacity region for a class of three-user partially connected interference channels. We focus on the impact of topology, interference alignment, and interplay between interference and noise. The representative channels we consider are the ones that have clear interference alignment gain. For these channels, Z-channel type outer bounds are tight to within a constant gap from capacity. We present near-optimal achievable schemes based on rate-splitting and lattice alignment.

cs.IT↗