Searcharxiv⌕ Search

arXiv subjects

Yifeng Mao

Publications and source records attributed to Yifeng Mao.

9 recordsLinked to original sources

Context as Infrastructure: Knowledge Compilation for Retrieval in Agentic Systems with Harnesses

Retrieval-augmented generation is becoming agentic. A search harness writes queries, reads results, reformulates, and decides when to stop. This largely solves accuracy: a harness with only BM25/grep and MiniMax M3 answers 85% of BrowseComp-Plus and 91% on multi-hop FRAMES in our runs, without compiled or vector indexes. It exposes a context problem: an undisciplined loop repeats queries, rereads documents, and fills the window with unused passages. The missing layer is knowledge compilation: a configurable ingestion pipeline where LLMs enrich documents with context no chunk contains alone--navigation hierarchies with claim-evidence leaves, entity graphs, timelines, mind maps, ontologies, wikis, and memory. Compilation is not a better retriever: under a competent harness, structural artifacts gave no detectable accuracy gain on FRAMES, and a PageIndex-style outline degraded at scale. It is a supply layer. A navigation tree whose leaves are claim-evidence pairs--the mechanism we measure--cuts query tokens by 91% when evidence suffices and adds 1.9% when it does not, a 38% average reduction; this implies a 2% break-even sufficiency rate and a measured rate near 43%. A decision gate cuts latency from 140 s to 30 s, though substitution's token saving is bought with retrieval rounds and carries its own latency cost. Other artifacts--ontology, graph, wiki, timeline, mind map, memory--pay off in deployment (governance, integration, onboarding, temporal orientation, continuity) where accuracy benchmarks cannot see them. We describe RAGFlow's pipeline and harness (open source at https://github.com/infiniflow/ragflow), which turns the loop into five disciplined stations and keeps the context entering the model the smallest sufficient set; together they form an agent-native database: a store that hands the agent compiled, governed context instead of a ranked list of passages.

cs.IR↗

Chemical transport by weakly nonlinear internal gravity waves in stars

While internal gravity waves (IGWs) are known to transport chemicals in the radiative zones of stars, substantial uncertainty remains regarding the magnitude and physical mechanisms of this transport. Most previous studies have relied on heuristic theories or numerical simulations that are difficult to extrapolate to stellar parameters. In this work, we present the first rigorous asymptotic calculation of (passive) chemical transport by IGWs, in the limit of small wave amplitude. The net vertical transport induced by a coherent packet of waves is found to be proportional to the horizontal mean of $(w^S+w^E)^2$, where $w^S$ and $w^E$ are the vertical Stokes and Eulerian mean velocities. The transport scales as the fourth power of wave amplitude and quadratically with time, and we verify these scalings with numerical simulations. In the weakly damped regime, the dominant transport coefficients scale as $ω^{-2}κ^2$, where $ω$ is the wave frequency and $κ$ is the radiative diffusivity. Because the transport can be positive or negative depending on the wave phases, the effective mixing induced by the superposition of coherent wave packets is expected to scale with the eighth power of wave amplitude. These results provide a theoretical foundation for making accurate predictions of IGW-induced chemical transport in stellar evolution models.

astro-ph.SR↗

FENAS: A Few-Shot Meta-Ensembling Framework for Efficient Neural Architecture Search with Synergized Zero-Cost Proxies

Efficient neural architecture search (NAS) requires accurate performance estimation under limited evaluation budgets. Existing zero-cost proxies (ZCPs) provide fast training-free signals, but each proxy usually captures only a partial property of an architecture, leading to unstable rankings across datasets and search spaces. In contrast, GNN-based predictors can improve accuracy, but often depend on labeled architectures, graph encoders, and costly representation learning, which makes them less suitable for extreme few-shot settings. We propose FENAS, a topology-agnostic few-shot prediction framework that turns complementary initialization priors into robust ranking signals. Its core component, CAST, is a feature pipeline that combines copula alignment and synergy transformation. CAST maps heterogeneous ZCPs into a compact feature manifold, enabling lightweight meta-learners to produce reliable rankings from scarce supervision. On NAS-Bench-201, with only 0.5% labeled samples, FENAS achieves a Kendall's Tau ($τ$) of 0.70, outperforming representative topology-based few-shot predictors in this low-label regime. In closed-domain search, FENAS evaluates the NAS-Bench-201 space in approximately 1138 seconds and achieves competitive test accuracy. By providing a fast, reliable, and structurally simple evaluation mechanism, FENAS supports the design of automated neural architectures under limited evaluation budgets.

cs.LG↗

PAG-Agent: Zero-Shot Personalized Music Album Generation via Aesthetic Profile Distillation

Large Music Models (LMMs) can synthesize high-fidelity audio and support increasingly flexible prompt control, yet modeling the subjective preferences of an individual listener remains underexplored. We present PAG-Agent, a zero-shot multimodal framework that uses an explicit Personal Aesthetic Profile to bridge user preference inference and music generation control. A multimodal LLM distills favorite-song audio and contextual user information into this structured representation, which a planning LLM translates into generation instructions for existing music synthesizers. We further introduce PA-Eval, which evaluates personalization through the Aesthetic Consistency Score (ACS) for target-user alignment and the Aesthetic Specificity Index (ASI) for cross-user specificity. Experiments with Personal-100 on multiple MLLM and music-generation backbones show consistent improvements over generic and direct conditioning. A 50-participant follow-up study further shows a correlation $r=0.64$ between ACS and subjective preference.

cs.MM↗

DeSIA: Attribute Inference Attacks Against Limited Fixed Aggregate Statistics

Empirical inference attacks are a popular approach for evaluating the privacy risk of data release mechanisms in practice. While an active attack literature exists to evaluate machine learning models or synthetic data release, we currently lack comparable methods for fixed aggregate statistics, in particular when only a limited number of statistics are released. We here propose an inference attack framework against fixed aggregate statistics and an attribute inference attack called DeSIA. We instantiate DeSIA against the U.S. Census PPMF dataset and show it to strongly outperform reconstruction-based attacks. In particular, we show DeSIA to be highly effective at identifying vulnerable users, achieving a true positive rate of 0.14 at a false positive rate of $10^{-3}$. We then show DeSIA to perform well against users whose attributes cannot be verified and when varying the number of aggregate statistics and level of noise addition. We also perform an extensive ablation study of DeSIA and show how DeSIA can be successfully adapted to the membership inference task. Overall, our results show that aggregation alone is not sufficient to protect privacy, even when a relatively small number of aggregates are being released, and emphasize the need for formal privacy mechanisms and testing before aggregate statistics are released.

cs.CR↗

Bright traveling breathers in media with long-range, nonconvex dispersion

The existence and properties of envelope solitary waves on a periodic, traveling wave background, called traveling breathers, are investigated numerically in representative nonlocal dispersive media. Using a fixed point computational scheme, a space-time boundary value problem for bright traveling breather solutions is solved for the weakly nonlinear Benjamin-Bona-Mahony equation, a nonlocal, regularized shallow water wave model, and the strongly nonlinear conduit equation, a nonlocal model of viscous core-annular flows. Curves of unit-mean traveling breather solutions within a three-dimensional parameter space are obtained. Resonance due to nonconvex, rational linear dispersion leads to a nonzero oscillatory background upon which traveling breathers propagate. These solutions exhibit a topological phase jump, so act as defects within the periodic background. For small amplitudes, traveling breathers are well-approximated by bright soliton solutions of the nonlinear Schrödinger equation with a negligibly small periodic background. These solutions are numerically continued into the large amplitude regime as elevation defects on cnoidal or cnoidal-like periodic traveling wave backgrounds. This study of bright traveling breathers provides insight into systems with nonconvex, nonlocal dispersion that occur in a variety of media such as internal oceanic waves subject to rotation and short, intense optical pulses.

nlin.PS↗

Observation of Traveling Breathers and Their Scattering in a Two-Fluid System

The observation of traveling breathers (TBs) with large-amplitude oscillatory tails realizes an almost 50-year-old theoretical prediction (Kuznetsov 1975) and generalizes the notion of a breather. Two strongly nonlinear TB families are created in a core-annular flow by interacting a soliton and a nonlinear periodic (cnoidal) carrier. Bright and dark TBs are observed to move faster or slower, respectively, than the carrier while imparting a phase shift. Agreement with model equations is achieved. Scattering of the TBs are observed to be physically elastic. The observed TBs generalize to many continuum and discrete systems.

nlin.PS↗

Long-time asymptotics and the radiation condition for linear evolution equations on the half-line with time-periodic boundary conditions

The large time $t$ asymptotics for scalar, constant coefficient,linear, third order, dispersive equations are obtained for asymptotically time-periodic Dirichlet boundary data and zero initial data on the half-line modeling a wavemaker acting upon an initially quiescent medium. The asymptotic Dirichlet-to-Neumann (D-N) map is constructed by expanding upon the recently developed $Q$-equation method. The D-N map is proven to be unique if and only if the radiation condition that selects the unique wavenumber branch of the dispersion relation for a sinusoidal, time-dependent boundary condition holds: (i) for frequencies in a finite interval, the wavenumber is real and corresponds to positive group velocity, (ii) for frequencies outside the interval, the wavenumber is complex with positive imaginary part. For fixed spatial location $x$, the corresponding asymptotic solution is (i) a traveling wave or (ii) a spatially decaying, time-periodic wave. Uniform-in-$x$ asymptotic solutions for the physical cases of the linearized Korteweg-de Vries and Benjamin-Bona-Mahony (BBM) equations are obtained via integral asymptotics. The linearized BBM asymptotics are found to quantitatively agree with viscous core-annular fluid experiments.

math.AP↗

Experimental investigations of linear and nonlinear periodic traveling waves in a viscous fluid conduit

Conduits generated by the buoyant dynamics between two miscible, Stokes fluids with high viscosity contrast exhibit rich nonlinear wave dynamics. However, little is known about the fundamental wave dispersion properties of the medium. In the present work, a pump is used to inject a time-periodic flow that results in the excitation of propagating small and large amplitude periodic traveling waves along the conduit interface. This wavemaker problem is used as a means to measure the linear and nonlinear dispersion relations and corresponding periodic traveling wave profiles. Measurements are favorably compared with predictions from a fully nonlinear, long-wave model (the conduit equation) and the analytically computed linear dispersion relation for two-Stokes flow. A critical frequency is observed, marking the threshold between propagating and non-propagating (spatially decaying) waves. Measurements of wave profiles and the wavenumber-frequency dispersion relation quantitatively agree with wave solutions of the conduit equation. An upshift from the conduit equation's predicted critical frequency is observed and is hypothesized to be explained by incorporating a weak recirculating flow into the full two-Stokes flow model which we observe to be operable in the experiment. When the boundary condition corresponds to the temporal profile of a nonlinear periodic traveling wave solution of the conduit equation, weakly nonlinear and strongly nonlinear, cnoidal-type waves are observed that quantitatively agree with the conduit nonlinear dispersion relation and wave profiles. This wavemaker problem is an important precursor to the experimental investigation of more general boundary value problems in viscous fluid conduit nonlinear wave dynamics.

physics.flu-dyn↗