arXiv subjects

Kevin Zhao

Publications and source records attributed to Kevin Zhao.

At least 19 recordsLinked to original sources

arXiv2026

Prediction-Guided Control in Data Center Networks

In this paper, we design, implement, and evaluate Polyphony, a system to give network operators a new way to control and reduce the frequency of poor tail latency events in multi-class data center networks, on the time scale of minutes. Polyphony is designed to be complementary to other adaptive mechanisms like congestion control and traffic engineering, but targets different aspects of network operation that have previously been considered static. By contrast to Polyphony, prior model-free optimization methods work best when there are only a few relevant degrees of freedom and where workloads and measurements are stable, assumptions not present in modern data center networks. Polyphony develops novel methods for measuring, predicting, and controlling network quality of service metrics for a dynamically changing workload. First, we monitor and aggregate workloads on a network-wide basis; we use the result as input to an approximate counterfactual prediction engine that estimates the effect of potential network configuration changes on network quality of service; we apply the best candidate and repeat in a closed-loop manner aimed at rapidly and stably converging to a configuration that meets operator goals. Using CloudLab on a simple topology, we observe that Polyphony converges to tight SLOs within ten minutes, and re-stabilizes after large workload shifts within fifteen minutes, while the prior state of the art fails to adapt.

cs.NI↗

arXiv2025

On zero-sum problems over metacyclic groups $C_n \rtimes_s C_2$

Let $G$ be a finite group. A finite collection of elements from $G$, where the order is disregarded and repetitions are allowed, is said to be a product-one sequence if its elements can be ordered such that their product in $G$ equals the identity element of $G$. Then, the Gao's constant $\mathsf E (G)$ of $G$ is the smallest integer $\ell$ such that every sequence of length at least $\ell$ has a product-one subsequence of length $|G|$. For a positive integer $n$, we denote by $C_n$ a cyclic group of order $n$. Let $G = C_n \rtimes_s C_2$ with $s^2\equiv 1\pmod n$ be a metacyclic group. The direct and inverse problems of $\mathsf E (G)$ were settled recently, except for the case that $G=C_{3n_2}\rtimes_s C_2$ with $n_2\neq 1$, $\gcd(n_2,6)=1$, $s\equiv -1 \pmod 3$, and $s\equiv 1\pmod {n_2}$. In this paper, we complete the remaining case and hence for all metacyclic groups of the form $G=C_n \rtimes C_2$, the Gao's constant and the associated inverse problem are now fully settled (see Theorem 1.2).

math.CO↗

arXiv2025

A Definition of AGI

The lack of a concrete definition for Artificial General Intelligence (AGI) obscures the gap between today's specialized AI and human-level cognition. This paper introduces a quantifiable framework to address this, defining AGI as matching the cognitive versatility and proficiency of a well-educated adult. To operationalize this, we ground our methodology in Cattell-Horn-Carroll theory, the most empirically validated model of human cognition. The framework dissects general intelligence into ten core cognitive domains-including reasoning, memory, and perception-and adapts established human psychometric batteries to evaluate AI systems. Application of this framework reveals a highly "jagged" cognitive profile in contemporary models. While proficient in knowledge-intensive domains, current AI systems have critical deficits in foundational cognitive machinery, particularly long-term memory storage. The resulting AGI scores (e.g., GPT-4 at 27%, GPT-5 at 57%) concretely quantify both rapid progress and the substantial gap remaining before AGI.

cs.AI↗

arXiv2025

On separating sets of polynomial invariants of finite abelian group actions

Let $G$ be a finite group acting on a finite dimensional complex vector space $V$ via linear transformations. Let $\mathbb{C}[V]^G$ be the algebra of polynomials that are invariant under the induced $G$-action on the polynomial ring $\mathbb{C}[V]$. A subset $S\subseteq\mathbb{C}[V]^G$ is a separating set if it separates the orbits of the group action. If $G$ is abelian, then there exist finite separating sets consisting of monomials. In this paper we investigate properties of separating sets from four different points of view, including the monoid theoretical properties of separating sets consisting of monomials, the minimal size of separating sets consisting of monomials, the exact value of the separating Noether number $\sepbeta(G)$ of abelian groups of rank $4$, and the inverse problem of $\sepbeta(G)$ for abelian groups of rank $2$.

math.AC↗

arXiv2025

Gravity-driven flux of particles through apertures

The gravity-driven discharge of granular material through an aperture is a fundamental problem in granular physics and is classically described by empirical laws with different fitting parameters. In this Letter, we disentangle the mass flux into distinct velocity and packing contributions by combining three-dimensional experiments and simulations. We define a dimensionless flux ratio that captures confinement-driven deviations from a free-fall limit, which is recovered when the aperture is large compared to the grain size. For spherical cohesionless grains, the deviations from the free-fall limit are captured by a single exponential correction factor over a characteristic length scale of $\sim$ 10-15 grain diameters. This is shown to be the scale over which the packing structure is modified due to the boundary. Building on the $\sqrt{gD}$ exit-velocity scaling, we propose a kinematic framework that explains the universality of granular discharge beyond empirical descriptions.

cond-mat.soft↗

arXiv2025

On zero-sum subsequences in a finite abelian group of length not exceeding a given number

Let $G$ be an additive finite abelian group and let $k\in [\exp(G),\mathsf{D}(G)-1]$ be a positive integer. Denote by $\mathsf{s}_{\leq k}(G)$ the smallest positive integer $l\in \mathbb{N}\cup \{+\infty\}$ such that each sequence of length $l$ over $G$ has a non-empty zero-sum subsequence of length at most $k$. Let $k_G\in [\exp(G),\mathsf{D}(G)-1]$ be the smallest positive integer such that $\mathsf{s}_{\leq \mathsf{D}(G)-d}(G)\leq \mathsf{D}(G)+d$ for $\mathsf{D}(G)-d\geq k_G$. We conjecture that $k_G=\frac{\mathsf{D}(G)+1}{2}$ for finite abelian groups $G$ with $r(G)\geq 2$ and $\mathsf{D}(G)=\mathsf{D}^*(G)$. In this paper, we mainly study this conjecture for finite abelian $p$-groups and get some results to support this conjecture. We also prove that $k_G\leq \mathsf{D}(G)-2$ for all finite abelian groups $G$ with $r(G)\geq 2$ except $C_2^3$ and $C_2^4$. In addition, we also get some lower bounds for the invariant $\mathsf{s}_{\leq k}(G)$.

math.CO↗

arXiv2025

On the separating Noether number of finite abelian groups

The separating Noether number $\beta_{\mathrm{sep}}(G)$ of a finite group $G$ is the minimal positive integer $d$ such that for every finite $G$-module $V$ there is a separating set consisting of invariant polynomials of degree at most $d$. In this paper we use methods from additive combinatorics to investigate the separating Noether number for finite abelian groups. Among others, we obtain the exact value of $\beta_{\mathrm{sep}}(G)$, provided that $G$ is either a $p$-group or has rank $2$, $3$ or $5$.

math.AC↗

arXiv2025

m4: A Learned Flow-level Network Simulator

Flow-level simulation is widely used to model large-scale data center networks due to its scalability. Unlike packet-level simulators that model individual packets, flow-level simulators abstract traffic as continuous flows with dynamically assigned transmission rates. While this abstraction enables orders-of-magnitude speedup, it is inaccurate by omitting critical packet-level effects such as queuing, congestion control, and retransmissions. We present m4, an accurate and scalable flow-level simulator that uses machine learning to learn the dynamics of the network of interest. At the core of m4 lies a novel ML architecture that decomposes state transition computations into distinct spatial and temporal components, each represented by a suitable neural network. To efficiently learn the underlying flow-level dynamics, m4 adds dense supervision signals by predicting intermediate network metrics such as remaining flow size and queue length during training. m4 achieves a speedup of up to 104$\times$ over packet-level simulation. Relative to a traditional flow-level simulation, m4 reduces per-flow estimation errors by 45.3% (mean) and 53.0% (p90). For closed-loop applications, m4 accurately predicts network throughput under various congestion control schemes and workloads.

cs.NI↗

arXiv2025

VeriPlan: Integrating Formal Verification and LLMs into End-User Planning

Automated planning is traditionally the domain of experts, utilized in fields like manufacturing and healthcare with the aid of expert planning tools. Recent advancements in LLMs have made planning more accessible to everyday users due to their potential to assist users with complex planning tasks. However, LLMs face several application challenges within end-user planning, including consistency, accuracy, and user trust issues. This paper introduces VeriPlan, a system that applies formal verification techniques, specifically model checking, to enhance the reliability and flexibility of LLMs for end-user planning. In addition to the LLM planner, VeriPlan includes three additional core features -- a rule translator, flexibility sliders, and a model checker -- that engage users in the verification process. Through a user study (n=12), we evaluate VeriPlan, demonstrating improvements in the perceived quality, usability, and user satisfaction of LLMs. Our work shows the effective integration of formal verification and user-control features with LLMs for end-user planning tasks.

cs.HC↗

arXiv2025

Learning Real-World Action-Video Dynamics with Heterogeneous Masked Autoregression

We propose Heterogeneous Masked Autoregression (HMA) for modeling action-video dynamics to generate high-quality data and evaluation in scaling robot learning. Building interactive video world models and policies for robotics is difficult due to the challenge of handling diverse settings while maintaining computational efficiency to run in real time. HMA uses heterogeneous pre-training from observations and action sequences across different robotic embodiments, domains, and tasks. HMA uses masked autoregression to generate quantized or soft tokens for video predictions. \ourshort achieves better visual fidelity and controllability than the previous robotic video generation models with 15 times faster speed in the real world. After post-training, this model can be used as a video simulator from low-level action inputs for evaluating policies and generating synthetic data. See this link https://liruiw.github.io/hma for more information.

cs.RO↗

arXiv2024

Dynamic and Compressive Adaptation of Transformers From Images to Videos

Recently, the remarkable success of pre-trained Vision Transformers (ViTs) from image-text matching has sparked an interest in image-to-video adaptation. However, most current approaches retain the full forward pass for each frame, leading to a high computation overhead for processing entire videos. In this paper, we present InTI, a novel approach for compressive image-to-video adaptation using dynamic Inter-frame Token Interpolation. InTI aims to softly preserve the informative tokens without disrupting their coherent spatiotemporal structure. Specifically, each token pair at identical positions within neighbor frames is linearly aggregated into a new token, where the aggregation weights are generated by a multi-scale context-aware network. In this way, the information of neighbor frames can be adaptively compressed in a point-by-point manner, thereby effectively reducing the number of processed frames by half each time. Importantly, InTI can be seamlessly integrated with existing adaptation methods, achieving strong performance without extra-complex design. On Kinetics-400, InTI reaches a top-1 accuracy of 87.1 with a remarkable 37.5% reduction in GFLOPs compared to naive adaptation. When combined with additional temporal modules, InTI achieves a top-1 accuracy of 87.6 with a 37% reduction in GFLOPs. Similar conclusions have been verified in other common datasets.

cs.CV↗

arXiv2024

VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding

Video Temporal Grounding (VTG) strives to accurately pinpoint event timestamps in a specific video using linguistic queries, significantly impacting downstream tasks like video browsing and editing. Unlike traditional task-specific models, Video Large Language Models (video LLMs) can handle multiple tasks concurrently in a zero-shot manner. Consequently, exploring the application of video LLMs for VTG tasks has become a burgeoning research area. However, despite considerable advancements in video content understanding, video LLMs often struggle to accurately pinpoint timestamps within videos, limiting their effectiveness in VTG tasks. To address this, we introduce VTG-LLM, a model designed to enhance video LLMs' timestamp localization abilities. Our approach includes: (1) effectively integrating timestamp knowledge into visual tokens; (2) incorporating absolute-time tokens to manage timestamp knowledge without concept shifts; and (3) introducing a lightweight, high-performance, slot-based token compression technique designed to accommodate the demands of a large number of frames to be sampled for VTG tasks. Additionally, we present VTG-IT-120K, a collection of publicly available VTG datasets that we have re-annotated to improve upon low-quality annotations. Our comprehensive experiments demonstrate the superior performance of VTG-LLM in comparison to other video LLM methods across a variety of VTG tasks.

cs.CV↗

arXiv2024

Omni-Recon: Harnessing Image-based Rendering for General-Purpose Neural Radiance Fields

Recent breakthroughs in Neural Radiance Fields (NeRFs) have sparked significant demand for their integration into real-world 3D applications. However, the varied functionalities required by different 3D applications often necessitate diverse NeRF models with various pipelines, leading to tedious NeRF training for each target task and cumbersome trial-and-error experiments. Drawing inspiration from the generalization capability and adaptability of emerging foundation models, our work aims to develop one general-purpose NeRF for handling diverse 3D tasks. We achieve this by proposing a framework called Omni-Recon, which is capable of (1) generalizable 3D reconstruction and zero-shot multitask scene understanding, and (2) adaptability to diverse downstream 3D applications such as real-time rendering and scene editing. Our key insight is that an image-based rendering pipeline, with accurate geometry and appearance estimation, can lift 2D image features into their 3D counterparts, thus extending widely explored 2D tasks to the 3D world in a generalizable manner. Specifically, our Omni-Recon features a general-purpose NeRF model using image-based rendering with two decoupled branches: one complex transformer-based branch that progressively fuses geometry and appearance features for accurate geometry estimation, and one lightweight branch for predicting blending weights of source views. This design achieves state-of-the-art (SOTA) generalizable 3D surface reconstruction quality with blending weights reusable across diverse tasks for zero-shot multitask scene understanding. In addition, it can enable real-time rendering after baking the complex geometry branch into meshes, swift adaptation to achieve SOTA generalizable 3D understanding performance, and seamless integration with 2D diffusion models for text-guided 3D editing.

cs.CV↗

arXiv2022

Life after BERT: What do Other Muppets Understand about Language?

Existing pre-trained transformer analysis works usually focus only on one or two model families at a time, overlooking the variability of the architecture and pre-training objectives. In our work, we utilize the oLMpics benchmark and psycholinguistic probing datasets for a diverse set of 29 models including T5, BART, and ALBERT. Additionally, we adapt the oLMpics zero-shot setup for autoregressive models and evaluate GPT networks of different sizes. Our findings show that none of these models can resolve compositional questions in a zero-shot fashion, suggesting that this skill is not learnable using existing pre-training objectives. Furthermore, we find that global model decisions such as architecture, directionality, size of the dataset, and pre-training objective are not predictive of a model's linguistic capabilities.

cs.CL↗

arXiv2022

Scalable Tail Latency Estimation for Data Center Networks

In this paper, we consider how to provide fast estimates of flow-level tail latency performance for very large scale data center networks. Network tail latency is often a crucial metric for cloud application performance that can be affected by a wide variety of factors, including network load, inter-rack traffic skew, traffic burstiness, flow size distributions, oversubscription, and topology asymmetry. Network simulators such as ns-3 and OMNeT++ can provide accurate answers, but are very hard to parallelize, taking hours or days to answer what if questions for a single configuration at even moderate scale. Recent work with MimicNet has shown how to use machine learning to improve simulation performance, but at a cost of including a long training step per configuration, and with assumptions about workload and topology uniformity that typically do not hold in practice. We address this gap by developing a set of techniques to provide fast performance estimates for large scale networks with general traffic matrices and topologies. A key step is to decompose the problem into a large number of parallel independent single-link simulations; we carefully combine these link-level simulations to produce accurate estimates of end-to-end flow level performance distributions for the entire network. Like MimicNet, we exploit symmetry where possible to gain additional speedups, but without relying on machine learning, so there is no training delay. On large-scale networks where ns-3 takes 11 to 27 hours to simulate five seconds of network behavior, our techniques run in one to two minutes with 99th percentile accuracy within 9% for flow completion times.

cs.NI↗

arXiv2022

Roomsemble: Progressive web application for intuitive property search

A successful real estate search process involves locating a property that meets a user's search criteria subject to an allocated budget and time constraints. Many studies have investigated modeling housing prices over time. However, little is known about how a user's tastes influence their real estate search and purchase decisions. It is unknown what house a user would choose taking into account an individual's personal tastes, behaviors, and constraints, and, therefore, creating an algorithm that finds the perfect match. In this paper, we investigate the first step in understanding a user's tastes by building a system to capture personal preferences. We concentrated our research on real estate photos, being inspired by house aesthetics, which often motivates prospective buyers into considering a property as a candidate for purchase. We designed a system that takes a user-provided photo representing that person's personal taste and recommends properties similar to the photo available on the market. The user can additionally filter the recommendations by budget and location when conducting a property search. The paper describes the application's overall layout including frontend design and backend processes for locating a desired property. The proposed model, which serves as the application's core, was tested with 25 users, and the study's findings, as well as some key conclusions, are detailed in this paper.

cs.IR↗

arXiv2021

Backpressure Flow Control

Effective congestion control for data center networks is becoming increasingly challenging with a growing amount of latency sensitive traffic, much fatter links, and extremely bursty traffic. Widely deployed algorithms, such as DCTCP and DCQCN, are still far from optimal in many plausible scenarios, particularly for tail latency. Many operators compensate by running their networks at low average utilization, dramatically increasing costs. In this paper, we argue that we have reached the practical limits of end-to-end congestion control. Instead, we propose, implement, and evaluate a new congestion control architecture called Backpressure Flow Control (BFC). BFC provides per-hop per-flow flow control, but with bounded state, constant-time switch operations, and careful use of buffers. We demonstrate BFC's feasibility by implementing it on Tofino2, a state-of-the-art P4-based programmable hardware switch. In simulation, we show that BFC achieves near optimal throughput and tail latency behavior even under challenging conditions such as high network load and incast cross traffic. Compared to existing end-to-end schemes, BFC achieves 2.3 - 60 X lower tail latency for short flows and 1.6 - 5 X better average completion time for long flows.

cs.NI↗

arXiv2021

Natural Adversarial Examples

We introduce two challenging datasets that reliably cause machine learning model performance to substantially degrade. The datasets are collected with a simple adversarial filtration technique to create datasets with limited spurious cues. Our datasets' real-world, unmodified examples transfer to various unseen models reliably, demonstrating that computer vision models have shared weaknesses. The first dataset is called ImageNet-A and is like the ImageNet test set, but it is far more challenging for existing models. We also curate an adversarial out-of-distribution detection dataset called ImageNet-O, which is the first out-of-distribution detection dataset created for ImageNet models. On ImageNet-A a DenseNet-121 obtains around 2% accuracy, an accuracy drop of approximately 90%, and its out-of-distribution detection performance on ImageNet-O is near random chance levels. We find that existing data augmentation techniques hardly boost performance, and using other public training datasets provides improvements that are limited. However, we find that improvements to computer vision architectures provide a promising path towards robust models.

cs.LG↗