arXiv subjects

Jiani Li

Publications and source records attributed to Jiani Li.

15 recordsLinked to original sources

arXiv2026

A Bio-Plausible Visual Neural Network for Locust-Inspired Collision Perception

Locust visual systems have long served as an important biological paradigm for studying looming perception and collision avoidance. Numerous computational models have successfully reproduced the selective responses of Lobula Giant Movement Detector (LGMD) neurons to approaching objects, thereby emulating the fundamental functionality of the biological system. However, existing models remain limited in biological plausibility and robustness when operating in complex and dynamic visual environments. To address these limitations, we propose a biologically plausible neural network for locust-inspired looming detection. The proposed framework incorporates a spatially isotropic sampling strategy that mimics the ommatidial organization of the locust compound eye, a population-voting mechanism inspired by population coding in biological neural systems, and leaky integrate-and-fire neuronal dynamics to replace conventional sigmoid-based membrane activation. Systematic experiments on synthetic stimuli, laboratory sequences, and real-world driving scenarios demonstrate that the proposed model improves robustness under challenging visual conditions while preserving computational efficiency and enhancing biological fidelity. These results highlight the potential of biologically grounded neural computation for robust and efficient collision perception.

cs.NE↗

arXiv2026

Exact-Penalty Prox-Linear Methods for Bilevel Optimization with $\ell_1$ Lower-Level Gradient Penalty

Bilevel optimization is a fundamental framework for hierarchical decision-making, but its solution is challenging due to the implicit and typically set-valued nature of the lower-level optimality condition. In this paper, we study bilevel optimization problems through an exact-penalty reformulation based on the $\ell_1$-norm of the lower-level gradient. Under suitable regularity assumptions, we show that this penalty defines a distance-bound function and yields an exact penalty property for sufficiently large penalty parameters. To solve the resulting nonsmooth penalized problem, we propose an exact-penalty prox-linear (EPPL) method and establish a stationarity-oriented convergence guarantee. We further specialize the method to the simple bilevel setting, where the subproblem admits an explicit dual reformulation as a box-constrained quadratic program. This structure leads to a dual spectral projected gradient method with closed-form primal recovery, for which convergence of the inner dual iterates is proved. Numerical experiments on a minimum-norm least-squares bilevel model show that the proposed method is effective in reducing both the lower-level and upper-level gaps to high accuracy. Compared with several existing methods, the proposed approach attains the best final solution accuracy on the tested instance.

math.OC↗

arXiv2026

RailVQA: A Benchmark and Framework for Efficient Interpretable Visual Cognition in Automatic Train Operation

As Automatic Train Operation (ATO) advances toward GoA4 and beyond, it increasingly depends on efficient, reliable cab-view visual perception and decision-oriented inference to ensure safe operation in complex and dynamic railway environments. However, existing approaches focus primarily on basic perception and often generalize poorly to rare yet safety-critical corner cases. They also lack the high-level reasoning and planning capabilities required for operational decision-making. Although recent Large Multi-modal Models (LMMs) show strong generalization and cognitive capabilities, their use in safety-critical ATO is hindered by high computational cost and hallucination risk. Meanwhile, reliable domain-specific benchmarks for systematically evaluating cognitive capabilities are still lacking. To address these gaps, we introduce RailVQA-bench, the first VQA benchmark for cab-view visual cognition in ATO, comprising 20,000 single-frame and 1,168 video based QA pairs to evaluate cognitive generalization and interpretability in both static and dynamic scenarios. Furthermore, we propose RailVQA-CoM, a collaborative large-small model framework that combines small-model efficiency with large-model cognition via a transparent three-module architecture and adaptive temporal sampling, improving perceptual generalization and enabling more efficient reasoning and planning. Experiments demonstrate that the proposed approach substantially improves performance, enhances interpretability, improves efficiency, and strengthens cross-domain generalization in autonomous driving systems. Code and datasets will be available at https://cybereye-bjtu.github.io/RailVQA.html.

cs.CV↗

arXiv2025

On Constraint Qualifications for MPECs with Applications to Bilevel Hyperparameter Optimization for Machine Learning

Constraint qualifications for a Mathematical Program with Equilibrium Constraints (MPEC) are essential for analyzing stationarity properties and establishing convergence results. In this paper, we explore several classical MPEC constraint qualifications and clarify the relationships among them. We subsequently examine the behavior of these constraint qualifications in the context of a specific MPEC derived from bilevel hyperparameter optimization (BHO) for L1-loss support vector classification. In particular, for such an MPEC, we provide a complete characterization of the well-known MPEC linear independence constraint qualification (MPEC-LICQ), therefore, establishing conditions under which it holds or fails for our BHO for support vector machines.

math.OC↗

arXiv2023

EDI: ESKF-based Disjoint Initialization for Visual-Inertial SLAM Systems

Visual-inertial initialization can be classified into joint and disjoint approaches. Joint approaches tackle both the visual and the inertial parameters together by aligning observations from feature-bearing points based on IMU integration then use a closed-form solution with visual and acceleration observations to find initial velocity and gravity. In contrast, disjoint approaches independently solve the Structure from Motion (SFM) problem and determine inertial parameters from up-to-scale camera poses obtained from pure monocular SLAM. However, previous disjoint methods have limitations, like assuming negligible acceleration bias impact or accurate rotation estimation by pure monocular SLAM. To address these issues, we propose EDI, a novel approach for fast, accurate, and robust visual-inertial initialization. Our method incorporates an Error-state Kalman Filter (ESKF) to estimate gyroscope bias and correct rotation estimates from monocular SLAM, overcoming dependence on pure monocular SLAM for rotation estimation. To estimate the scale factor without prior information, we offer a closed-form solution for initial velocity, scale, gravity, and acceleration bias estimation. To address gravity and acceleration bias coupling, we introduce weights in the linear least-squares equations, ensuring acceleration bias observability and handling outliers. Extensive evaluation on the EuRoC dataset shows that our method achieves an average scale error of 5.8% in less than 3 seconds, outperforming other state-of-the-art disjoint visual-inertial initialization approaches, even in challenging environments and with artificial noise corruption.

cs.RO↗

arXiv2022

Ferrimagnetism in stable non-metal covalent organic framework

We synthesized a pure organic non-metal crystalline covalent organic framework TAPA-BTD-COF by bottom-up Schiff base chemical reaction. And this imine-based COF is stable in aerobic condition and room-temperature. We discovered that this TAPA-BTD-COF exhibited strong magneticity in 300 K generating magnetic hysteresis loop in M-H characterization and giant chimol up to 0.028. And we further conducted zero-field cooling and field-cooling measurement of M-T curves. The as-synthesized materials showed a large chi/mol up to 0.028 in 300 K and increasing to 0.037 in 4.0 K with 200 Oe measurement field. The TAPA-BTD-COF 1/chimol~T curve supported its ferrimagnetism, with an intrinsic delta temperature as -33.03 K by extrapolating the 1/chimol~T curve. From the continuously increasing slope of 1/chimol~T, we consider that this TAPA-BTD-COF belongs to ferrimagnetic other than antiferromagnetic materials. And the large chimol value 0.028 at 300 K and 0.037 at 4.0 K also supported this, since common antiferromagnetic materials possess chimol in the range of 10-5 to 10-3 as weak magnetics other than strong magnetic materials such as ferrimagnetics and ferromagnetics. Since this material is purely non-metal organic polymer, the possibility of d-block and f-block metal with unpaired-electron induced magnetism can be excluded. Besides, since the COF does not involve free-radical monomer in the processes of synthesis, we can also exclude the origin of free-radical induced magnetism. According to recent emerging flat-band strong correlated exotic electron property, this unconventional phenomenon may relate to n-type doping on the flat-band locating in the CBM, thus generating highly-localized electron with infinite effective mass and exhibiting strong correlation, which accounts for this non-trivial strong and stable ferrimagneticity at room-temperature and aerobic atmospheric conditions.

cond-mat.mtrl-sci↗

arXiv2021

EDDA: Explanation-driven Data Augmentation to Improve Explanation Faithfulness

Recent years have seen the introduction of a range of methods for post-hoc explainability of image classifier predictions. However, these post-hoc explanations may not always be faithful to classifier predictions, which poses a significant challenge when attempting to debug models based on such explanations. To this end, we seek a methodology that can improve the faithfulness of an explanation method with respect to model predictions which does not require ground truth explanations. We achieve this through a novel explanation-driven data augmentation (EDDA) technique that augments the training data with occlusions inferred from model explanations; this is based on the simple motivating principle that \emph{if} the explainer is faithful to the model \emph{then} occluding salient regions for the model prediction should decrease the model confidence in the prediction, while occluding non-salient regions should not change the prediction. To verify that the proposed augmentation method has the potential to improve faithfulness, we evaluate EDDA using a variety of datasets and classification models. We demonstrate empirically that our approach leads to a significant increase of faithfulness, which can facilitate better debugging and successful deployment of image classification models in real-world applications.

cs.LG↗

arXiv2020

Byzantine Resilient Distributed Multi-Task Learning

Distributed multi-task learning provides significant advantages in multi-agent networks with heterogeneous data sources where agents aim to learn distinct but correlated models simultaneously.However, distributed algorithms for learning relatedness among tasks are not resilient in the presence of Byzantine agents. In this paper, we present an approach for Byzantine resilient distributed multi-task learning. We propose an efficient online weight assignment rule by measuring the accumulated loss using an agent's data and its neighbors' models. A small accumulated loss indicates a large similarity between the two tasks. In order to ensure the Byzantine resilience of the aggregation at a normal agent, we introduce a step for filtering out larger losses. We analyze the approach for convex models and show that normal agents converge resiliently towards the global minimum.Further, aggregation with the proposed weight assignment rule always results in an improved expected regret than the non-cooperative case. Finally, we demonstrate the approach using three case studies, including regression and classification problems, and show that our method exhibits good empirical performance for non-convex models, such as convolutional neural networks.

cs.LG↗

arXiv2020

Interplay Between Resilience and Accuracy in Resilient Vector Consensus in Multi-Agent Networks

In this paper, we study the relationship between resilience and accuracy in the resilient distributed multi-dimensional consensus problem. We consider a network of agents, each of which has a state in $\mathbb{R}^d$. Some agents in the network are adversarial and can change their states arbitrarily. The normal (non-adversarial) agents interact locally and update their states to achieve consensus at some point in the convex hull $\calC$ of their initial states. This objective is achievable if the number of adversaries in the neighborhood of normal agents is less than a specific value, which is a function of the local connectivity and the state dimension $d$. However, to be resilient against adversaries, especially in the case of large $d$, the desired local connectivity is large. We discuss that resilience against adversarial agents can be improved if normal agents are allowed to converge in a bounded region $\calB\supseteq\calC$, which means normal agents converge at some point close to but not necessarily inside $\calC$ in the worst case. The accuracy of resilient consensus can be measured by the Hausdorff distance between $\calB$ and $\calC$. As a result, resilience can be improved at the cost of accuracy. We propose a resilient bounded consensus algorithm that exploits the trade-off between resilience and accuracy by projecting $d$-dimensional states into lower dimensions and then solving instances of resilient consensus in lower dimensions. We analyze the algorithm, present various resilience and accuracy bounds, and also numerically evaluate our results.

eess.SY↗

arXiv2020

Resilient Distributed Diffusion for Multi-task Estimation

Distributed diffusion is a powerful algorithm for multi-task state estimation which enables networked agents to interact with neighbors to process input data and diffuse information across the network. Compared to a centralized approach, diffusion offers multiple advantages that include robustness to node and link failures. In this paper, we consider distributed diffusion for multi-task estimation where networked agents must estimate distinct but correlated states of interest by processing streaming data. By exploiting the adaptive weights used for diffusing information, we develop attack models that drive normal agents to converge to states selected by the attacker. The attack models can be used for both stationary and non-stationary state estimation. In addition, we develop a resilient distributed diffusion algorithm under the assumption that the number of compromised nodes in the neighborhood of each normal node is bounded by $F$ and we show that resilience may be obtained at the cost of performance degradation. Finally, we evaluate the proposed attack models and resilient distributed diffusion algorithm using stationary and non-stationary multi-target localization.

cs.MA↗

arXiv2020

Resilient Distributed Diffusion in Networks with Adversaries

In this paper, we study resilient distributed diffusion for multi-task estimation in the presence of adversaries where networked agents must estimate distinct but correlated states of interest by processing streaming data. We show that in general diffusion strategies are not resilient to malicious agents that do not adhere to the diffusion-based information processing rules. In particular, by exploiting the adaptive weights used for diffusing information, we develop time-dependent attack models that drive normal agents to converge to states selected by the attacker. We show that an attacker that has complete knowledge of the system can always drive its targeted agents to its desired estimates. Moreover, an attacker that does not have complete knowledge of the system including streaming data of targeted agents or the parameters they use in diffusion algorithms, can still be successful in deploying an attack by approximating the needed information. The attack models can be used for both stationary and non-stationary state estimation.In addition, we present and analyze a resilient distributed diffusion algorithm that is resilient to any data falsification attack in which the number of compromised agents in the local neighborhood of a normal agent is bounded. The proposed algorithm guarantees that all normal agents converge to their true target states if appropriate parameters are selected. We also analyze trade-off between the resilience of distributed diffusion and its performance in terms of steady-state mean-square-deviation (MSD) from the correct estimates. Finally, we evaluate the proposed attack models and resilient distributed diffusion algorithm using stationary and non-stationary multi-target localization.

cs.MA↗

arXiv2020

Detecting Adversarial Examples in Learning-Enabled Cyber-Physical Systems using Variational Autoencoder for Regression

Learning-enabled components (LECs) are widely used in cyber-physical systems (CPS) since they can handle the uncertainty and variability of the environment and increase the level of autonomy. However, it has been shown that LECs such as deep neural networks (DNN) are not robust and adversarial examples can cause the model to make a false prediction. The paper considers the problem of efficiently detecting adversarial examples in LECs used for regression in CPS. The proposed approach is based on inductive conformal prediction and uses a regression model based on variational autoencoder. The architecture allows to take into consideration both the input and the neural network prediction for detecting adversarial, and more generally, out-of-distribution examples. We demonstrate the method using an advanced emergency braking system implemented in an open source simulator for self-driving cars where a DNN is used to estimate the distance to an obstacle. The simulation results show that the method can effectively detect adversarial examples with a short detection delay.

cs.LG↗

arXiv2020

Resilient Distributed Vector Consensus Using Centerpoints

In this paper, we study the resilient vector consensus problem in networks with adversarial agents and improve resilience guarantees of existing algorithms. A common approach to achieving resilient vector consensus is that every non-adversarial (or normal) agent in the network updates its state by moving towards a point in the convex hull of its \emph{normal} neighbors' states. Since an agent cannot distinguish between its normal and adversarial neighbors, computing such a point, often called as \emph{safe point}, is a challenging task. To compute a safe point, we propose to use the notion of \emph{centerpoint}, which is an extension of the median in higher dimensions, instead of Tverberg partition of points, which is often used for this purpose. We discuss that the notion of centerpoint provides a complete characterization of safe points in $\mathbb{R}^d$. In particular, we show that a safe point is essentially an interior centerpoint if the number of adversaries in the neighborhood of a normal agent $i$ is less than $\frac{N_i}{d+1} $, where $d$ is the dimension of the state vector and $N_i$ is the total number of agents in the neighborhood of $i$. Consequently, we obtain necessary and sufficient conditions on the number of adversarial agents to guarantee resilient vector consensus. Further, by considering the complexity of computing centerpoints, we discuss improvements in the resilience guarantees of vector consensus algorithms and compare with the other existing approaches. Finally, we numerically evaluate the performance of our approach through experiments.

eess.SY↗

arXiv2020

Accurate Temporal Action Proposal Generation with Relation-Aware Pyramid Network

Accurate temporal action proposals play an important role in detecting actions from untrimmed videos. The existing approaches have difficulties in capturing global contextual information and simultaneously localizing actions with different durations. To this end, we propose a Relation-aware pyramid Network (RapNet) to generate highly accurate temporal action proposals. In RapNet, a novel relation-aware module is introduced to exploit bi-directional long-range relations between local features for context distilling. This embedded module enhances the RapNet in terms of its multi-granularity temporal proposal generation ability, given predefined anchor boxes. We further introduce a two-stage adjustment scheme to refine the proposal boundaries and measure their confidence in containing an action with snippet-level actionness. Extensive experiments on the challenging ActivityNet and THUMOS14 benchmarks demonstrate our RapNet generates superior accurate proposals over the existing state-of-the-art methods.

cs.CV↗

arXiv2019

Relation-Aware Pyramid Network (RapNet) for temporal action proposal

In this technical report, we describe our solution to temporal action proposal (task 1) in ActivityNet Challenge 2019. First, we fine-tune a ResNet-50-C3D CNN on ActivityNet v1.3 based on Kinetics pretrained model to extract snippet-level video representations and then we design a Relation-Aware Pyramid Network (RapNet) to generate temporal multiscale proposals with confidence score. After that, we employ a two-stage snippet-level boundary adjustment scheme to re-rank the order of generated proposals. Ensemble methods are also been used to improve the performance of our solution, which helps us achieve 2nd place.

cs.CV↗