arXiv subjects

Yifei Zhu

Publications and source records attributed to Yifei Zhu.

At least 19 recordsLinked to original sources

arXiv2026

Scalable Context Orchestration for Serving LLMs Over Voice

Voice AI applications are gaining popularity as advances in large language models (LLMs) enable more natural and accessible spoken interactions. Serving these applications requires accounting not only for what users say, but also for how they speak (e.g., speaking rate) and the conditions under which their audio is captured and transmitted (e.g., background noise and packet loss). However, existing LLM systems represent conversation context as a flat, growing sequence of messages, leaving voice-specific context implicit in the audio. As a result, they can generate responses that are poorly aligned with user preferences, degrade interaction quality under adverse environmental conditions, and incur high costs over long voice sessions. We present llmovoice, a context-management middleware that explicitly models voice context and orchestrates its use. At each turn, llmovoice constructs a bounded voice context from the current user input, relevant interaction history, and explicit paralinguistic and environmental states. It then uses the serving LLM to reason over this context and generate runtime directives that guide how the system responds. We evaluate llmovoice on real-world voice applications and benchmarks. It reduces speaking-rate alignment error by 52.4%, lowers the false-interruption rate from 46.0% to 0.9% under packet loss, and reduces model usage cost by 79.2%. For long sessions, llmovoice reduces per-turn cost by up to 24.9 times while retaining up to 98.7% of baseline answer quality.

cs.SD↗

arXiv2026

MoRAE: Flow-Friendly Self-Supervised Latents for Text-to-Motion Generation

Text-to-motion generation must produce motions that are semantically correct, temporally coherent, and physically plausible. A natural approach is to first project motion data into a structured semantic space and then train a generative model within that space. Such a paradigm has been highly successful in image generation through Representation Autoencoders (RAEs), where a frozen self-supervised encoder provides semantic features for diffusion or flow models to learn from. However, direct transfer of such a paradigm to motion space using Motion-JEPA as the frozen encoder fails dramatically. We diagnose this failure geometrically and identify two motion-specific bottlenecks: (1) the JEPA feature space is spectrally ill-conditioned, making the Gaussian-to-data transport unstable; and (2) even with a well-conditioned spectrum, flow residuals tend to align with decoder-sensitive directions, where small latent errors are amplified into large motion artifacts after decoding. Based on these insights, we propose MoRAE. MoRAE addresses the two bottlenecks separately. A compact bottleneck distills the structured JEPA representation while removing weak and redundant directions, bringing the latent spectrum into a transport-stable regime. Motion-coupled training then aligns the retained latent geometry with the decoder, making characteristic flow errors less costly after decoding. With this flow-friendly latent, a standard non-autoregressive Flow-Matching DiT achieves state-of-the-art performance.

cs.CV↗

arXiv2026

A quantum chemistry dataset containing ground-state and conical-intersection structures of 260k molecules

Conical intersections play central roles in photoinduced reactions. However, comprehensive conical-intersection datasets that could advance our understanding of excited-state reaction processes remain scarce. To address this gap, we constructed a quantum chemistry dataset containing ground-state and conical-intersection structures of small molecules (up to ten heavy atoms: C, N, O, F). Ground-state geometries were optimized at the semi-empirical OM2 level, with single-point energies calculated at the OM2/MRCI level. Conical-intersection geometries and energies were also computed at the OM2/MRCI level. This dataset is designed to enable a deep integration of photochemistry with machine learning, bridging the gap between photochemical insight and data-driven approaches.

physics.chem-ph↗

arXiv2026

PolitNuggets: Benchmarking Agentic Discovery of Long-Tail Political Facts

Large Reasoning Models (LRMs) embedded in agentic frameworks have transformed information retrieval from static, long context question answering into open-ended exploration. Yet real world use requires models to discover and synthesize "long-tail" facts from dispersed sources, a capability that remains under-evaluated. We introduce PolitNuggets, a multilingual benchmark for agentic information synthesis via constructing political biographies for 400 global elites, covering over 10000 political facts. We standardize evaluation with an optimized multi agent system and propose FactNet, an evidence conditional protocol that scores discovery, fine-grained accuracy, and efficiency. Across models and settings, we find that current systems often struggle with fine-grained details, and vary substantially in efficiency. Finally, using benchmark diagnostics, we relate agent performance to underlying model capabilities, highlighting the importance of short-context extraction, multilingual robustness, and reliable tool use.

cs.AI↗

arXiv2026

Agentic Framework for Political Biography Extraction

The production of large-scale political datasets typically demands extracting structured facts from vast piles of unstructured documents or web sources, a task that traditionally relies on expensive human experts and remains prohibitively difficult to automate at scale. In this paper, we leverage Large Language Models (LLMs) to automate the extraction of multi-dimensional elite biographies, addressing a long-standing bottleneck in political science research. We propose a two-stage ``Synthesis-Coding'' framework for complex extraction task: an upstream synthesis stage that uses recursive agentic LLMs to search, filter, and curate biography from heterogeneous web sources, followed by a downstream coding stage that maps curated biography into structured dataframes. We validate this framework through three primary results. First, we demonstrate that, when given curated contexts, LLM coders match or outperform human experts in extraction accuracy. Second, we show that in web environments, the agentic system synthesizes more information from web resources than human collective intelligence (Wikipedia). Finally, we diagnosed that directly coding from long and multi-language corpora introduces bias that the synthesis stage can alleviate by curating evidence into signal-dense representations. By comprehensive evaluation, We provide a generalizable, scalable framework for building transparent and expansible large scale database in political science.

cs.CL↗

arXiv2026

ViTMAlis: Towards Latency-Critical Mobile Video Analytics with Vision Transformers

Edge-assisted mobile video analytics (MVA) applications are increasingly shifting from using vision models based on convolutional neural networks (CNNs) to those built on vision transformers (ViTs) to leverage their superior global context modeling and generalization capabilities. However, deploying these advanced models in latency-critical MVA scenarios presents significant challenges. Unlike traditional CNN-based offloading paradigms where network transmission is the primary bottleneck, ViT-based systems are constrained by substantial inference delays, particularly for dense prediction tasks where the need for high-resolution inputs exacerbates the inherent quadratic computational complexity of ViTs. To address these challenges, we propose a dynamic mixed-resolution inference strategy tailored for ViT-backboned dense prediction models, enabling flexible runtime trade-offs between speed and accuracy. Building on this, we introduce ViTMAlis, a ViT-native device-to-edge offloading framework that dynamically adapts to network conditions and video content to jointly reduce transmission and inference delays. We implement a fully functional prototype of ViTMAlis on commodity mobile and edge devices. Extensive experiments demonstrate that, compared to state-of-the-art accuracy-centric, content-aware, and latency-adaptive baselines, ViTMAlis significantly reduces end-to-end offloading latency while improving user-perceived rendering accuracy, providing a practical foundation for next-generation mobile intelligence.

cs.NI↗

arXiv2025

Bridging Visual Intuition and Chemical Expertise: An Autonomous Analysis Framework for Nonadiabatic Dynamics Simulations via Mentor-Engineer-Student Collaboration

Analyzing nonadiabatic molecular dynamics trajectories traditionally heavily relies on expert intuition and visual pattern recognition, a process that is difficult to formalize. We present VisU, a vision-driven framework that leverages the complementary strengths of two state-of-the-art large language models to establish a "virtual research collective." This collective operates through a "Mentor-Engineer-Student" paradigm that mimics the collaborative intelligence of a professional chemistry laboratory. Within this ecosystem, the Mentor provides physical intuition through visual reasoning, while the Engineer adaptively constructs analysis scripts, and the Student executes the pipeline and manages the data and results. VisU autonomously orchestrates a four-stage workflow comprising Preprocessing, Recursive Channel Discovery, Important-Motion Identification, and Validation/Summary. This systematic approach identifies reaction channels and key nuclear motions while generating professional academic reports. By bridging visual insight with chemical expertise, VisU establishes a new paradigm for human-AI collaboration in the analysis of excited-state dynamics simulation results, significantly reducing dependence on manual interpretation and enabling more intuitive, scalable mechanistic discovery.

physics.chem-ph↗

arXiv2025

Hyperion: Low-Latency Ultra-HD Video Analytics via Collaborative Vision Transformer Inference

Recent advancements in array-camera videography enable real-time capturing of ultra-high-definition (Ultra-HD) videos, providing rich visual information in a large field of view. However, promptly processing such data using state-of-the-art transformer-based vision foundation models faces significant computational overhead in on-device computing or transmission overhead in cloud computing. In this paper, we present Hyperion, the first cloud-device collaborative framework that enables low-latency inference on Ultra-HD vision data using off-the-shelf vision transformers over dynamic networks. Hyperion addresses the computational and transmission bottleneck of Ultra-HD vision transformers by exploiting the intrinsic property in vision Transformer models. Specifically, Hyperion integrates a collaboration-aware importance scorer that identifies critical regions at the patch level, a dynamic scheduler that adaptively adjusts patch transmission quality to balance latency and accuracy under dynamic network conditions, and a weighted ensembler that fuses edge and cloud results to improve accuracy. Experimental results demonstrate that Hyperion enhances frame processing rate by up to 1.61 times and improves the accuracy by up to 20.2% when compared with state-of-the-art baselines under various network environments.

cs.DC↗

arXiv2025

AirGS: Real-Time 4D Gaussian Streaming for Free-Viewpoint Video Experiences

Free-viewpoint video (FVV) enables immersive viewing experiences by allowing users to view scenes from arbitrary perspectives. As a prominent reconstruction technique for FVV generation, 4D Gaussian Splatting (4DGS) models dynamic scenes with time-varying 3D Gaussian ellipsoids and achieves high-quality rendering via fast rasterization. However, existing 4DGS approaches suffer from quality degradation over long sequences and impose substantial bandwidth and storage overhead, limiting their applicability in real-time and wide-scale deployments. Therefore, we present AirGS, a streaming-optimized 4DGS framework that rearchitects the training and delivery pipeline to enable high-quality, low-latency FVV experiences. AirGS converts Gaussian video streams into multi-channel 2D formats and intelligently identifies keyframes to enhance frame reconstruction quality. It further combines temporal coherence with inflation loss to reduce training time and representation size. To support communication-efficient transmission, AirGS models 4DGS delivery as an integer linear programming problem and design a lightweight pruning level selection algorithm to adaptively prune the Gaussian updates to be transmitted, balancing reconstruction quality and bandwidth consumption. Extensive experiments demonstrate that AirGS reduces quality deviation in PSNR by more than 20% when scene changes, maintains frame-level PSNR consistently above 30, accelerates training by 6 times, reduces per-frame transmission size by nearly 50% compared to the SOTA 4DGS approaches.

cs.GR↗

arXiv2025

An Automated Framework for Analyzing Structural Evolution in On-the-fly Non-adiabatic Molecular Dynamics Using Autoencoder and Multiple Molecular Descriptors

A major challenge in nonadiabatic molecular dynamics is to automatically and objectively identify the key reaction coordinates that drive molecules toward distinct excited-state decay channels. Traditional manual analyses are inefficient and rely heavily on expert intuition, creating a bottleneck for interpreting complex photochemical processes. To overcome this, we introduce a fully automated machine-learning framework that directly extracts these coordinates from on-the-fly trajectory surface hopping data. By combining an Autoencoder for nonlinear dimensionality reduction with clustering and information entropy analysis, our method autonomously maps reaction channels and pinpoints their governing structural motions. When applied to keto isocytosine and the methaniminium cation, the framework objectively revealed invovled reaction channels and corresponding active coordinates with high efficiency and accuracy. This work establishes an effective paradigm for mechanistic insight in excited-state dynamics, transforming raw trajectory data into clear, interpretable reaction mechanisms.

physics.chem-ph↗

arXiv2025

Chemistry-Enhanced Diffusion-Based Framework for Small-to-Large Molecular Conformation Generation

Obtaining 3D conformations of realistic polyatomic molecules at the quantum chemistry level remains challenging, and although recent machine learning advances offer promise, predicting large-molecule structures still requires substantial computational effort. Here, we introduce StoL, a diffusion model-based framework that enables rapid and knowledge-free generation of large molecular structures from small-molecule data. Remarkably, StoL assembles molecules in a LEGO-style fashion from scratch, without seeing the target molecules or any structures of comparable size during training. Given a SMILES input, it decomposes the molecule into chemically valid fragments, generates their 3D structures with a diffusion model trained on small molecules, and assembles them into diverse conformations. This fragment-based strategy eliminates the need for large-molecule training data while maintaining high scalability and transferability. By embedding chemical principles into key steps, StoL ensures faster convergence, chemically rational structures, and broad configurational coverage, as confirmed against DFT calculations.

physics.chem-ph↗

arXiv2025

LAFA: Agentic LLM-Driven Federated Analytics over Decentralized Data Sources

Large Language Models (LLMs) have shown great promise in automating data analytics tasks by interpreting natural language queries and generating multi-operation execution plans. However, existing LLM-agent-based analytics frameworks operate under the assumption of centralized data access, offering little to no privacy protection. In contrast, federated analytics (FA) enables privacy-preserving computation across distributed data sources, but lacks support for natural language input and requires structured, machine-readable queries. In this work, we present LAFA, the first system that integrates LLM-agent-based data analytics with FA. LAFA introduces a hierarchical multi-agent architecture that accepts natural language queries and transforms them into optimized, executable FA workflows. A coarse-grained planner first decomposes complex queries into sub-queries, while a fine-grained planner maps each subquery into a Directed Acyclic Graph of FA operations using prior structural knowledge. To improve execution efficiency, an optimizer agent rewrites and merges multiple DAGs, eliminating redundant operations and minimizing computational and communicational overhead. Our experiments demonstrate that LAFA consistently outperforms baseline prompting strategies by achieving higher execution plan success rates and reducing resource-intensive FA operations by a substantial margin. This work establishes a practical foundation for privacy-preserving, LLM-driven analytics that supports natural language input in the FA setting.

cs.AI↗

arXiv2025

Neural Network Based Framework for Passive Intermodulation Cancellation in MIMO Systems

Passive intermodulation (PIM) has emerged as a critical source of self-interference in modern MIMO-OFDM systems, especially under the stringent requirements of 5G and beyond. Conventional cancellation methods often rely on complex nonlinear models with limited scalability and high computational cost. In this work, we propose a lightweight deep learning framework for PIM cancellation that leverages depthwise separable convolutions and dilated convolutions to efficiently capture nonlinear dependencies across antennas and subcarriers. To further enhance convergence, we adopt a cyclic learning rate schedule and gradient clipping. In a controlled MIMO experimental setup, the method effectively suppresses third-order passive intermodulation (PIM) distortion, achieving up to 29dB of average power error (APE) with only 11k trainable parameters. These results highlight the potential of compact neural architectures for scalable interference mitigation in future wireless communication systems.

eess.SP↗

arXiv2025

Spectral moduli problems for level structures and an integral Jacquet-Langlands dual of Morava E-theory

Given an E-infinity ring spectrum R, with motivation from chromatic homotopy theory, we define relative effective Cartier divisors for a spectral Deligne-Mumford stack over Spet(R) and prove that, as a functor from connective R-algebras to topological spaces, it is representable. This enables us to solve various moduli problems of level structures on spectral abelian varieties, overcoming difficulty at primes dividing the level. In particular, we obtain higher-homotopical refinement for finite levels of a Lubin-Tate tower as E-infinity ring spectra, which generalizes Morava, Hopkins, Miller, Goerss, and Lurie's spectral realization of the deformation ring at the ground level. Moreover, passing to the infinite level and then descending along the equivariantly isomorphic Drinfeld tower, we obtain a Jacquet-Langlands dual to the Morava E-theory spectrum, along with homotopy fixed point spectral sequences dual to those studied by Devinatz and Hopkins. These serve as potential tools for computing higher-periodic homotopy types from pro-etale cohomology of p-adic general linear groups.

math.AT↗

arXiv2025

OmniSense: Towards Edge-Assisted Online Analytics for 360-Degree Videos

With the reduced hardware costs of omnidirectional cameras and the proliferation of various extended reality applications, more and more $360^\circ$ videos are being captured. To fully unleash their potential, advanced video analytics is expected to extract actionable insights and situational knowledge without blind spots from the videos. In this paper, we present OmniSense, a novel edge-assisted framework for online immersive video analytics. OmniSense achieves both low latency and high accuracy, combating the significant computation and network resource challenges of analyzing $360^\circ$ videos. Motivated by our measurement insights into $360^\circ$ videos, OmniSense introduces a lightweight spherical region of interest (SRoI) prediction algorithm to prune redundant information in $360^\circ$ frames. Incorporating the video content and network dynamics, it then smartly scales vision models to analyze the predicted SRoIs with optimized resource utilization. We implement a prototype of OmniSense with commodity devices and evaluate it on diverse real-world collected $360^\circ$ videos. Extensive evaluation results show that compared to resource-agnostic baselines, it improves the accuracy by $19.8\%$ -- $114.6\%$ with similar end-to-end latencies. Meanwhile, it hits $2.0\times$ -- $2.4\times$ speedups while keeping the accuracy on par with the highest accuracy of baselines.

cs.NI↗

arXiv2025

Efficient Serving of LLM Applications with Probabilistic Demand Modeling

Applications based on Large Language Models (LLMs) contains a series of tasks to address real-world problems with boosted capability, which have dynamic demand volumes on diverse backends. Existing serving systems treat the resource demands of LLM applications as a blackbox, compromising end-to-end efficiency due to improper queuing order and backend warm up latency. We find that the resource demands of LLM applications can be modeled in a general and accurate manner with Probabilistic Demand Graph (PDGraph). We then propose Hermes, which leverages PDGraph for efficient serving of LLM applications. Confronting probabilistic demand description, Hermes applies the Gittins policy to determine the scheduling order that can minimize the average application completion time. It also uses the PDGraph model to help prewarm cold backends at proper moments. Experiments with diverse LLM applications confirm that Hermes can effectively improve the application serving efficiency, reducing the average completion time by over 70% and the P95 completion time by over 80%.

cs.DC↗

arXiv2025

B2LoRa: Boosting LoRa Transmission for Satellite-IoT Systems with Blind Coherent Combining

With the rapid growth of Low Earth Orbit (LEO) satellite networks, satellite-IoT systems using the LoRa technique have been increasingly deployed to provide widespread Internet services to low-power and low-cost ground devices. However, the long transmission distance and adverse environments from IoT satellites to ground devices pose a huge challenge to link reliability, as evidenced by the measurement results based on our real-world setup. In this paper, we propose a blind coherent combining design named B2LoRa to boost LoRa transmission performance. The intuition behind B2LoRa is to leverage the repeated broadcasting mechanism inherent in satellite-IoT systems to achieve coherent combining under the low-power and low-cost constraints, where each re-transmission at different times is regarded as the same packet transmitted from different antenna elements within an antenna array. Then, the problem is translated into aligning these packets at a fine granularity despite the time, frequency, and phase offsets between packets in the case of frequent packet loss. To overcome this challenge, we present three designs - joint packet sniffing, frequency shift alignment, and phase drift mitigation to deal with ultra-low SNRs and Doppler shifts featured in satellite-IoT systems, respectively. Finally, experiment results based on our real-world deployments demonstrate the high efficiency of B2LoRa.

cs.NI↗

arXiv2025

$\gamma$-FedHT: Stepsize-Aware Hard-Threshold Gradient Compression in Federated Learning

Gradient compression can effectively alleviate communication bottlenecks in Federated Learning (FL). Contemporary state-of-the-art sparse compressors, such as Top-$k$, exhibit high computational complexity, up to $\mathcal{O}(d\log_2{k})$, where $d$ is the number of model parameters. The hard-threshold compressor, which simply transmits elements with absolute values higher than a fixed threshold, is thus proposed to reduce the complexity to $\mathcal{O}(d)$. However, the hard-threshold compression causes accuracy degradation in FL, where the datasets are non-IID and the stepsize $\gamma$ is decreasing for model convergence. The decaying stepsize reduces the updates and causes the compression ratio of the hard-threshold compression to drop rapidly to an aggressive ratio. At or below this ratio, the model accuracy has been observed to degrade severely. To address this, we propose $\gamma$-FedHT, a stepsize-aware low-cost compressor with Error-Feedback to guarantee convergence. Given that the traditional theoretical framework of FL does not consider Error-Feedback, we introduce the fundamental conversation of Error-Feedback. We prove that $\gamma$-FedHT has the convergence rate of $\mathcal{O}(\frac{1}{T})$ ($T$ representing total training iterations) under $\mu$-strongly convex cases and $\mathcal{O}(\frac{1}{\sqrt{T}})$ under non-convex cases, \textit{same as FedAVG}. Extensive experiments demonstrate that $\gamma$-FedHT improves accuracy by up to $7.42\%$ over Top-$k$ under equal communication traffic on various non-IID image datasets.

cs.LG↗