arXiv subjects

Jiaqi Lu

Publications and source records attributed to Jiaqi Lu.

13 recordsLinked to original sources

arXiv2026

Characterizing Bias in Post-Bandit Inference under Index Algorithms

Bandit algorithms generate data for downstream inference, but adaptive sampling biases post-bandit sample means. We analyze this bias for stable index algorithms, including UCB1 and its generalizations, and derive sharp leading-order expressions for the sample-mean bias and expected $Z$-statistic. Our characterization reveals the algorithmic origin of bias through a key index-function-dependent quantity, which we term effective exploration rate. For example, under UCB1, the effective exploration rate is of order $\sqrt{\log T}$, and the standardized bias of any arm (that is not uniquely optimal) decays at the extremely slow rate $1/\sqrt{\log T}$. We also show how the choice of the index function affects both regret and bias, which reveals a regret-bias trade-off: more exploratory algorithm reduces bias but increases regret. Our sharp characterization for bias uses a novel empirical fluid approximation of the algorithm's sampling dynamics, which may be of independent interest.

cs.LG↗

arXiv2026

Count Bridges enable Modeling and Deconvolving Transcriptomic Data

Many modern biological assays, including RNA sequencing, yield integer-valued counts that reflect the number of molecules detected. These measurements are often not at the desired resolution: while the unit of interest is typically a single cell, many measurement technologies produce counts aggregated over sets of cells. Although recent generative frameworks such as diffusion and flow matching have been extended to non-Euclidean and discrete settings, it remains unclear how best to model integer-valued data or how to systematically deconvolve aggregated observations. We introduce Count Bridges, a stochastic bridge process on the integers that provides an exact, tractable analogue of diffusion-style models for count data, with closed-form conditionals for efficient training and sampling. We extend this framework to enable direct training from aggregated measurements via an Expectation-Maximization-style approach that treats unit-level counts as latent variables. We demonstrate state-of-the-art performance on integer distribution matching benchmarks, comparing against flow matching and discrete flow matching baselines across various metrics. We then apply Count Bridges to two large-scale problems in biology: modeling single-cell gene expression data at the nucleotide resolution, with applications to deconvolving bulk RNA-seq, and resolving multicellular spatial transcriptomic spots into single-cell count profiles. Our methods offer a principled foundation for generative modeling and deconvolution of biological count data across scales and modalities.

cs.LG↗

arXiv2026

Bandit Allocational Instability

When multi-armed bandit (MAB) algorithms allocate pulls among competing arms, the resulting allocation can exhibit huge variation. This is particularly harmful in modern applications such as learning-enhanced platform operations and post-bandit statistical inference. Thus motivated, we introduce a new performance metric of MAB algorithms termed allocation variability, which is the largest (over arms) standard deviation of an arm's number of pulls. We establish a fundamental trade-off between allocation variability and regret, the canonical performance metric of reward maximization. In particular, for any algorithm, the worst-case regret $R_T$ and worst-case allocation variability $S_T$ must satisfy $R_T \cdot S_T=\Omega(T^{\frac{3}{2}})$ as $T\rightarrow\infty$, as long as $R_T=o(T)$. This indicates that any minimax regret-optimal algorithm must incur worst-case allocation variability $\Theta(T)$, the largest possible scale; while any algorithm with sublinear worst-case regret must necessarily incur ${S}_T= \omega(\sqrt{T})$. We further show that this lower bound is essentially tight, and that any point on the Pareto frontier $R_T \cdot S_T=\tilde{\Theta}(T^{3/2})$ can be achieved by a simple tunable algorithm UCB-f, a generalization of the classic UCB1. Finally, we discuss implications for platform operations and for statistical inference, when bandit algorithms are used. As a byproduct of our result, we resolve an open question of Praharaj and Khamaru (2025).

cs.LG↗

arXiv2026

ES-MemEval: Benchmarking Conversational Agents on Personalized Long-Term Emotional Support

Large Language Models (LLMs) have shown strong potential as conversational agents. Yet, their effectiveness remains limited by deficiencies in robust long-term memory, particularly in complex, long-term web-based services such as online emotional support. However, existing long-term dialogue benchmarks primarily focus on static and explicit fact retrieval, failing to evaluate agents in critical scenarios where user information is dispersed, implicit, and continuously evolving. To address this gap, we introduce ES-MemEval, a comprehensive benchmark that systematically evaluates five core memory capabilities: information extraction, temporal reasoning, conflict detection, abstention, and user modeling, in long-term emotional support settings, covering question answering, summarization, and dialogue generation tasks. To support the benchmark, we also propose EvoEmo, a multi-session dataset for personalized long-term emotional support that captures fragmented, implicit user disclosures and evolving user states. Extensive experiments on open-source long-context, commercial, and retrieval-augmented (RAG) LLMs show that explicit long-term memory is essential for reducing hallucinations and enabling effective personalization. At the same time, RAG improves factual consistency but struggles with temporal dynamics and evolving user states. These findings highlight both the potential and limitations of current paradigms and motivate more robust integration of memory and retrieval for long-term personalized dialogue systems.

cs.CL↗

arXiv2025

AC^0[p]-Frege Cannot Efficiently Prove that Constant-Depth Algebraic Circuit Lower Bounds are Hard

We study whether lower bounds against constant-depth algebraic circuits computing the Permanent over finite fields (Limaye-Srinivasan-Tavenas, J. ACM 2025; Forbes, CCC 2024) are hard to prove in certain proof systems. We focus on a DNF formula that expresses that such lower bounds are hard for constant-depth algebraic proofs. Using an adaptation of the diagonalization framework of Santhanam and Tzameret (SIAM J. Comput. 2025), we show unconditionally that this family of DNF formulas does not admit polynomial-size propositional AC0[p]-Frege proofs infinitely often. This rules out the possibility that the DNF family is easy, and establishes that its status is either that of a hard tautology for AC0[p]-Frege or else unprovable (not a tautology). While it remains open whether the DNFs in question are tautologies, we provide evidence in this direction. In particular, under the plausible assumption that certain weak properties of multilinear algebra, specifically those involving tensor rank, do not admit short constant-depth algebraic proofs, the DNFs are tautologies. We also observe that several weaker variants of the DNF formula are provably tautologies, and we show that the question of whether the DNFs are tautologies connects to conjectures of Razborov (ICALP 1996) and Krajicek (J. Symb. Log. 2004). Our result has two additional features. (i) Existential depth amplification: the DNF formula is parameterised by a constant depth d bounding the depth of the algebraic proofs. We show that there exists some fixed depth d such that if there are no small depth-d algebraic proofs of certain circuit lower bounds for the Permanent, then there are no such small algebraic proofs in any constant depth. (ii) Necessity: we show that our result is a necessary step towards establishing lower bounds against constant-depth algebraic proofs, and more generally against any sufficiently strong proof system.

cs.CC↗

arXiv2025

A Surrogate model for High Temperature Superconducting Magnets to Predict Current Distribution with Neural Network

Finite element methods (FEM) for high-temperature superconducting (HTS) magnets become time-consuming at larger scales, restricting the rapid optimization of meter-scale REBCO solenoids. In this work, a surrogate model based on a fully connected residual neural network (FCRN) is developed to predict the current density distribution in REBCO solenoids. Trained on datasets generated from FEM simulations by the T-A formulation, the FCRN model is evaluated under both fast ramping and steady-state scenarios, showing a lower validation loss than the fully connected network (FCN). When extrapolating geometric parameters beyond the training set, the model achieves a relative error of below 10 % for magnetization losses in Case 1 and an average error of 1.2 % for the central magnetic field in Case 2. Furthermore, deploying the steady-state surrogate model for rapid magnet design found the optimal solution within the parameter space under constraints, with a relative central magnetic field error of 0.2 % compared to FEM results. With rapid predictions, this surrogate model offers an efficient tool for the intelligent design of large-scale HTS magnets.

cs.LG↗

arXiv2025

Lower Bounds against the Ideal Proof System in Finite Fields

Lower bounds against strong algebraic proof systems and specifically fragments of the Ideal Proof System (IPS), have been obtained in an ongoing line of work. All of these bounds, however, are proved only over large (or characteristic $0$) fields, yet finite fields are the more natural setting for propositional proof complexity, especially for progress toward lower bounds for Frege systems such as $AC^0[p]$-Frege. This work establishes lower bounds against fragments of IPS over fixed finite fields. Specifically, we show that a variant of the knapsack instance studied by Govindasamy, Hakoniemi, and Tzameret (FOCS'22) has no polynomial-size IPS refutation over finite fields when the refutation is multilinear and written as a constant-depth circuit. The key ingredient of our argument is the recent set-multilinearization result of Forbes (CCC'24), which extends the earlier result of Limaye, Srinivasan, and Tavenas (FOCS'21) to all fields, and an extension of the techniques of Govindasamy, Hakoniemi, and Tzameret to finite fields. We also separate this proof system from the one studied by Govindasamy, Hakoniemi, and Tzameret. In addition, we present new lower bounds for read-once algebraic branching program refutations, roABP-IPS, in finite fields, extending results of Forbes, Shpilka, Tzameret, and Wigderson (Theor. of Comput.'21) and Hakoniemi, Limaye, and Tzameret (STOC'24). Finally, we show that any lower bound against any proof system at least as strong as (non-multilinear) constant-depth IPS over finite fields for any instance, even a purely algebraic instance (i.e., not a translation of a Boolean formula or CNF), implies a hard CNF formula for the respective IPS fragment, and hence an $AC^0[p]$-Frege lower bound by known simulations over finite fields (Grochow and Pitassi (J. ACM'18)).

cs.CC↗

arXiv2025

Intelligent Joint Security and Delay Determinacy Performance Guarantee Strategy in RIS-Assisted IIoT Communication Systems

With the advancement of the Industrial Internet of Things (IIoT), IIoT services now exhibit diverse Quality of Service (QoS) requirements in terms of delay, determinacy, and security, which pose significant challenges for alignment with existing network resources. Reconfigurable Intelligent Surface (RIS), a key enabling technology for IIoT, not only optimizes signal propagation and enhances network performance but also ensures secure communication and deterministic delays to mitigate threats such as data leakage and eavesdropping. In this paper, we conduct a deterministic delay analysis under a specified decoding error rate for RIS-assisted IIoT communication systems using Stochastic Network Calculus (SNC). We propose an on-demand joint strategy to maximize delay determinacy while guaranteeing secure transmission performance. This is achieved by jointly optimizing the transmit power, channel blocklength (CBL) at the user end, and the phase shift matrix at the RIS. Furthermore, we introduce a State Interdependence-Driven Parameterized Deep Q-Network (SID-PDQN) algorithm to intelligently enforce on-demand performance guarantees. Simulation results demonstrate that the proposed SID-PDQN algorithm significantly enhances network performance compared to baseline methods such as DQN, Dueling-DQN, and DDPG.

eess.SP↗

arXiv2025

A characterization of sample adaptivity in UCB data

We characterize a joint CLT of the number of pulls and the sample mean reward of the arms in a stochastic two-armed bandit environment under UCB algorithms. Several implications of this result are in place: (1) a nonstandard CLT of the number of pulls hence pseudo-regret that smoothly interpolates between a standard form in the large arm gap regime and a slow-concentration form in the small arm gap regime, and (2) a heuristic derivation of the sample bias up to its leading order from the correlation between the number of pulls and sample means. Our analysis framework is based on a novel perturbation analysis, which is of broader interest on its own.

stat.ML↗

arXiv2024

Tetragonal BaCoO$_3$: A Co$^{4+}$ Ferromagnetic Mott Insulator with Inverted Spin Crossover

The interplay between crystal electric field splitting of d states and Hund's rule exchange energy in cobalt-based perovskites offers a promising avenue for inducing spin-state transitions. This study reports a new body-centered tetragonal (BCT) phase of BaCoO$_3$ (BCT-BaCoO$_3$), synthesized under high pressure (15 GPa) and high temperature (1200 {\deg}C) conditions. BCT-BaCoO$_3$ adopts a double perovskite structure of EuTiO$_3$-type (space group I4/mcm, #140), confirmed by high-resolution scanning transmission electron microscopy. X-ray photoelectron spectroscopy reveals a rare Co$^{4+}$ valence state. Magnetization and X-ray absorption measurements reveal a low-spin to high-spin transition that takes place between 200 and 300 K. While spin crossovers are relatively common among common oxides, the one observed in BCT-BaCoO$_3$ is remarkable in that it proceeds in the opposite direction from conventional spin transitions. BCT-BaCoO$_3$ exhibits a low-spin (S = 1/2) state at high temperatures and transitions to a high-spin (S = 5/2) state at low temperatures. Within the high-spin state, hard ferromagnetic order onsets at T$_C$ = 107 K. Electrical resistivity indicates weak magnetoresistance and insulating behavior. Overall, BCT-BaCoO$_3$ presents an exceptional model for the exploration of spin-state transitions and the study of Co spin states in cobalt-based perovskites.

cond-mat.mtrl-sci↗

arXiv2024

Design of a large-scale superconducting dipole magnet for the CEE spectrometer

The CSR External-target Experiment (CEE) is a large-scale spectrometer under construction at the Heavy Ion Research Facility in Lanzhou (HIRFL) for studying the phase structure of nuclear matter at high baryon density and the equation of states of nuclear matter at supra-saturation densities. One of the key components is a large acceptance dipole magnet with a central field of 0.5 T and the homogeneity of 5% within a 1 m long, 1.2 m wide, and 0.9 m high aperture. Detectors will be installed within this aperture. An innovative design for the superconducting detector magnet is proposed that goes beyond the conventional approach. The magnet is designed as a coil-dominant type, with conductors discretized on a racetrack-shaped cross-section to generate the necessary fields. A warm iron yoke is used to enhance the central field and minimize the stray field. The magnet has overall dimensions of 3.4 meters in length, 2.7 meters in height, and 4.3 meters in width. The coils will be wound using a 19-strand rope cable comprised of 12 NbTi superconducting wires and 7 copper wires. The ratio of copper to superconductor of the cable is 6.9. The keel supports serve as the primary structural support for the coils to withstand the electromagnetic force. The coils will be indirectly cooled by liquid helium within three external helium vessels. To ensure reliable protection of the magnet during a quench, an active protection method combined with quench-back effect is employed. In this paper, we mainly present the detailed design of the magnetic field, structure, quench protection and cryostat for the spectrometer magnet.

physics.ins-det↗

arXiv2023

Effective electrical manipulation of topological antiferromagnet by orbital Hall effect

Electrical control of the non-trivial topology in Weyl antiferromagnet is of great interests to develop next-generation spintronic devices. Recent works suggest that spin Hall effect can switch the topological antiferromagnetic order. However, the switching efficiency remains relatively low. Here, we demonstrate effective manipulation of antiferromagnetic order in Weyl semimetal Mn3Sn by orbital Hall effect originated from metal Mn or oxide CuOx. While Mn3Sn is proven to be able to convert orbit current to spin current by itself, we find that inserting a heavy metal layer like Pt with proper thickness can effectively reduce the critical switching current density by one order of magnitude. In addition, we show that the memristor-like switching behavior of Mn3Sn can mimic the potentiation and depression processes of a synapse with high linearity, which is beneficial for constructing artificial neural network with high accuracy. Our work paves an alternative way to manipulate topological antiferromagnetic order and may inspire more high-performance antiferromagnetic functional devices.

cond-mat.mtrl-sci↗

arXiv2023

SOT-MRAM-Enabled Probabilistic Binary Neural Networks for Noise-Tolerant and Fast Training

We report the use of spin-orbit torque (SOT) magnetoresistive random-access memory (MRAM) to implement a probabilistic binary neural network (PBNN) for resource-saving applications. The in-plane magnetized SOT (i-SOT) MRAM not only enables field-free magnetization switching with high endurance (> 10^11), but also hosts multiple stable probabilistic states with a low device-to-device variation (< 6.35%). Accordingly, the proposed PBNN outperforms other neural networks by achieving an 18* increase in training speed, while maintaining an accuracy above 97% under the write and read noise perturbations. Furthermore, by applying the binarization process with an additional SOT-MRAM dummy module, we demonstrate an on-chip MNIST inference performance close to the ideal baseline using our SOT-PBNN hardware.

physics.app-ph↗