Federated fine-tuning efficiently adapts large pre-trained models to new tasks by using additional data while minimizing re-training costs. This approach enhances data privacy and reduces computational demands but relies on a central server, often cloud-based, which is vulnerable to adversarial attacks that can compromise the aggregation process. We propose${\sf FLAGuard}$, a novel and efficient verification scheme that specifically addresses these challenges in federated Low-Rank Adaptation (LoRA) settings.${\sf FLAGuard}$is the first to introduce a two-stage verification process specifically designed for LoRA-based aggregation. In the first stage, the scheme independently verifies the correctness of the aggregated$A$and$B$matrices. In the second stage, it verifies the multiplication result of the aggregated$A$and$B$matrices, ensuring the correctness of the final LoRA parameters. Additionally, we introduce the Iterative Gradient Sampling and Convolutional Compression (IGSCC) technique, which combines probabilistic sampling with convolutional operations to efficiently reduce the dimensionality of gradient matrices. This enables secure verification without sacrificing model performance. Our comprehensive security analysis of${\sf FLAGuard}$further establishes its reliability in federated learning environments. Extensive experimental results demonstrate that${\sf FLAGuard}$achieves over a$100\times$speedup in the aggregation verification phase and reduces communication overhead by more than 50% compared to state-of-the-art methods.

The performance of pedestrian detectors typically relies on sufficient labeled data, and semi-supervised learning is a promising way to address the deficiency in manual annotations by utilizing sufficient unlabeled images. In this work, we design a Structure-Controllable Pedestrian Instance Generation approach (SCPIG), which is tailored to semi-supervised pedestrian detection. Specifically, we adopt a mask encoder to transform mask images into the embeddings encapsulating structure knowledge. In addition, we incorporate a mapping network to transform random latent code and a conditional generation network to synthesize diverse pedestrian instances, where the transformed code and mask embedding control pedestrian appearance and structure, respectively. The synthesized pedestrian instances are used to construct high-quality pseudo-labeled images for training pedestrian detectors. Extensive experiments validate the effectiveness of SCPIG in controllable pedestrian instance synthesizing and semi-supervised pedestrian detection.

A multi-zone (typically 8×8) time-of-flight (ToF) sensor offers a low-cost, low-power, and compact solution for range measurement, making it ideal for specialized robotic applications. However, its low resolution limits its usability. Pairing a ToF sensor with a camera enhances depth perception and can solve the unscaled metric problem in mono depth estimation. Advances in deep learning further enable high-quality depth map reconstruction from ToF-camera data, providing a cost-effective alternative. However, accurate ToF-camera calibration remains a challenge due to ToF sensor’s coarse depth output.This work presents a simple yet effective method for the extrinsic calibration of a ToF sensor with an RGB camera using only a chessboard and two whiteboards. A tailored two-plane fitting algorithm is proposed specifically for the ToF sensor. Moreover, our approach leverages parallel lines with vanishing points and geometric constraints from plane intersections. This eliminates the need for robotic arm movements or SLAM-based sensor pose reconstruction, significantly reducing complexity while maintaining high accuracy. Experimental results demonstrate that our method lowers the root mean square (RMS) depth difference from 96.59 mm to 67.89 mm, underscoring its effectiveness in practical applications. Code is publicly available in https://github.com/Tianyou-Nottingham/ToF-Camera-Calibration.

Federated Learning (FL) represents a distributed machine learning approach, enabling the joint training of a global model through the aggregation of gradients from participating clients without necessitating the exchange of raw data. Prior research has explored methods for verifying the correctness of aggregation in this context and mitigating the overhead associated with the verification process. Nonetheless, the advent of Large Language Models (LLMs), with their parameters numbering in the billions, presents ongoing challenges in devising efficient verification mechanisms in FL for large models. In this paper, we propose an innovative Efficient Verifiable Federated Learning scheme${\sf LaVFL}$, which focusing on addressing the verification challenges incurred by LLM. Specifically, we propose an efficient layer-by-layer verification approach for LLMs by designing a Convolution Gradient Compression (CGC) method without compromising model accuracy. Additionally, to minimize computational and communication overheads, we propose an efficient verification strategy PGS, namely, a Probabilistic Gradient Sampling strategy, which aims to reduce the gradient dimensions for each round of verification while ensuring a high probability of comprehensive verification. We implement a prototype of${\sf LaVFL}$, and extensive experimental results demonstrate that${\sf LaVFL}$achieves over a$300 \times$speedup in the aggregation verification phase and reduces communication overheads by more than 75%, compared to VeriFL under the same experimental setup.

Oriented object detection has witnessed significant progress in recent years. However, the impressive performance of oriented object detectors is at the huge cost of labor-intensive annotations, and deteriorates once the an-notated data becomes limited. Semi-supervised learning, in which sufficient unannotated data are utilized to enhance the base detector, is a promising method to address the annotation deficiency problem. Motivated by weakly supervised learning, we introduce annotation-efficient point annotations for unannotated images and propose a weakly semi-supervised method for oriented object detection to balance the detection performance and annotation cost. Specifically, we propose a Rotation-Modulated Relational Graph Matching method to match relations of proposals centered on an-notated points between the teacher and student models to alleviate the ambiguity of point annotations in depicting the oriented object. In addition, we further propose a Relational Rank Distribution Matching method to align the rank distribution on classification and regression between different models. Finally, to handle the difficult annotated points that both models are confused about, we introduce weakly supervised learning to impose positive signals for difficult point-induced clusters to the base model, and focus the base model on the occupancy between the predictions and an-notated points. We perform extensive experiments on chal-lenging datasets to demonstrate the effectiveness of our proposed weakly semi-supervised method in leveraging point-annotated data for significant performance improvement.

In today's big data era, there is a tremendously huge amount of data available. Layman users lack not only the knowledge and experience in data analytics to make sense of these data but also the computational resources for executing the analytics. In this paper, we propose and develop a layman analytics system (LAS), which provides the layman users with a scalable and ready-to-use analytics tool to automatically generate analytics workflows for classification tasks. The LAS is designed to benefit from existing open-source data analytics tools using generic ontological modeling of analytics operators from these tools as well as adaptive constraint refinement for metadata learning. Moreover, the LAS can be deployed on both public and private clouds to cater to the need of scalable computing and easy maintenance. To demonstrate the performance of the LAS, we conducted experiments with 114 data sets obtained from the University of California Irvine Machine Learning Repository. The workflows generated by the LAS were benchmarked against the OpenML whereby each data set has a range of classification accuracy obtained using classifiers designed and fine-tuned by data experts. The comparisons showed that 87 out of 114 data sets have exceeded the 50th percentile of the benchmark data. Among these 87 data sets, the LAS outperforms the 90th percentile of the benchmarks on 49 data sets.

Formulation of appropriate data analytics workflows requires intricate knowledge and rich experiences of data analytics experts. This problem is further compounded by continuous advancement and improvement in analytical algorithms. In this paper, a generic non-domain specific solution for the creation of appropriate workflows targeted at supervised learning problems is proposed. Our adaptive workflow recommendation engine based on collaborative analytics matches analytics needs with relevant workflows in repository. It is capable of picking workflows with better performance as compared to randomly selected workflows. The recommendation engine is now augmented by a workflow optimizer that applies genetic programming to further improve the recommended workflows through iterative evolution, leading to better alternative workflows. This unique Collaborative Analytics Recommender System is tested on seven UCI benchmark datasets. It is shown that the final workflows produced by the system could closely approximate, in terms of accuracy, the best workflows that analytics experts could possibly design.

There are many ways to build a predictive model from data. Besides the numerous classification or regression algorithms to choose from, there are countless possibilities of useful data transformation prior to modeling. To assist in discovering good predictive analytics workflows, we introduced recently a collaborative analytics system that allows workflow sharing and reuse. We designed a recommendation engine for the system to enable matching of analytics needs with relevant workflows stored in repository. The engine relies on meta-predictive modeling of traffic-analysis workflow-characteristics. In this paper, we present a feasibility study of applying this collaborative analytics system to predict traffic congestion. Different ways to build predictive models from traffic dataset are pooled as shared workflows. We demonstrate that through dynamic recommendation of workflows that are suitable for the real-time varying traffic data, a reliable congestion prediction can be achieved. The promising results showcase that systematic collaboration among data scientists made possible by our system can be a powerful tool to produce very accurate prediction from data.

Supervised learning is well-known and widely applied in many domains including bioinformatics, cheminformatics and financial forecasting. However, the interference from irrelevant features may lead to the poor accuracy of classifiers. As a popular feature selection model, GA-SVM is desirable in many of those cases to filter out irrelevant features and improve the learning performance subsequently. However, the high computational cost strongly discourages the application of GA-SVM in large-scale datasets. In this paper, an HPC-enabled GA-SVM (HGA-SVM) is proposed by integrating data parallelization, multithreading and heuristic techniques with the ultimate goal of robustness and low computational cost. Our proposed model is comprised of four improvement strategies: 1) GA parallelization, 2) SVM parallelization, 3) neighbor search and 4) evaluation caching. All the four strategies improve various aspects of the feature selection model and contribute collectively towards higher computational throughput.