Point cloud-based 3D single object tracking (3D SOT) plays a pivotal role in applications such as autonomous driving and robotic vision. Despite recent progress, most existing approaches rely solely on current-frame features for target localization. This approach overlooks temporal information that is crucial for robust tracking under occlusion, appearance variations, and sparse point clouds. In addition, the effectiveness of 3D SOT largely depends on the quality of feature fusion between the target template and the search region. Traditional fusion strategies often suffer from limited interaction capacity and weak discriminative representation. To address these challenges, we propose DT-Tracker, which performs multi-layer bidirectional feature interaction and temporal cue propagation to improve tracking robustness and feature discrimination capability. Specifically, we introduce a Dual-Context Propagation Network that applies bidirectional cross-attention across multiple layers between the template and search region, enabling deep semantic alignment and progressive feature refinement. Furthermore, we design a Temporal Context Fusion module that adaptively incorporates temporal cues from historical fusion features into the current frame, effectively improving resilience to occlusion and appearance drift. Extensive experiments on the KITTI and nuScenes datasets demonstrate that DT-Tracker achieves competitive results compared to existing representative methods.

Instruction tuning is critical for adapting large language models (LLMs) to downstream tasks, and recent studies have demonstrated that small amounts of human-curated data can outperform larger datasets, challenging traditional data scaling laws. While LLM-based data quality rating systems offer a cost-effective alternative to human annotation, they often suffer from inaccuracies and biases, even in powerful models like GPT-4. In this work, we introduce $DS^2$, a **D**iversity-aware **S**core curation method for **D**ata **S**election. By systematically modeling error patterns through a score transition matrix, $DS^2$ corrects LLM-based scores and promotes diversity in the selected data samples. Our approach shows that a curated subset (just 3.3\% of the original dataset) outperforms full-scale datasets (300k samples) across various machine-alignment benchmarks, and matches or surpasses human-aligned datasets such as LIMA with the same sample size (1k samples). These findings challenge conventional data scaling assumptions, highlighting that redundant, low-quality samples can degrade performance and reaffirming that ``more can be less''.

Unlearning in Large Language Models (LLMs) is essential for ensuring ethical and responsible AI use, especially in addressing privacy leak, bias, safety, and evolving regulations. Existing approaches to LLM unlearning often rely on retain data or a reference LLM, yet they struggle to adequately balance unlearning performance with overall model utility. This challenge arises because leveraging explicit retain data or implicit knowledge of retain data from a reference LLM to fine-tune the model tends to blur the boundaries between the forgotten and retain data, as different queries often elicit similar responses. In this work, we propose eliminating the need to retain data or the reference LLM for response calibration in LLM unlearning. Recognizing that directly applying gradient ascent on the forget data often leads to optimization instability and poor performance, our method guides the LLM on what not to respond to, and importantly, how to respond, based on the forget data. Hence, we introduce Forget data only Loss AjustmenT (FLAT), a "flat" loss adjustment approach which addresses these issues by maximizing $f$-divergence between the available template answer and the forget answer only w.r.t. the forget data. The variational form of the defined $f$-divergence theoretically provides a way of loss adjustment by assigning different importance weights for the learning w.r.t. template responses and the forgetting of responses subject to unlearning. Empirical results demonstrate that our approach not only achieves superior unlearning performance compared to existing methods but also minimizes the impact on the model’s retained capabilities, ensuring high utility across diverse tasks, including copyrighted content unlearning on Harry Potter dataset and MUSE Benchmark, and entity unlearning on the TOFU dataset.

Recent studies show that in supervised fine-tuning (SFT) of large language models (LLMs), data quality matters more than quantity. While most data cleaning methods concentrate on filtering entire samples, the quality of individual tokens within a sample can vary significantly. After pre-training, even in high-quality samples, patterns or phrases that are not task-related can be redundant, uninformative, or even harmful. Continuing to fine-tune on these patterns may offer limited benefit and even degrade downstream task performance. In this paper, we investigate token quality from a noisy-label perspective and propose a generic token cleaning pipeline for SFT tasks. Our method filters out uninformative tokens while preserving those carrying key task-specific information. Specifically, we first evaluate token quality by examining the influence of model updates on each token, then apply a threshold-based separation. The token influence can be measured in a single pass with a fixed reference model or iteratively with self-evolving reference models. The benefits and limitations of both methods are analyzed theoretically by error upper bounds. Extensive experiments show that our framework consistently improves downstream performance. Code is available at https://github.com/UCSC-REAL/TokenCleaning.

The recent success of generative AI highlights the crucial role of high-quality human feedback in building trustworthy AI systems. However, the increasing use of large language models (LLMs) by crowdsourcing workers poses a significant challenge: datasets intended to reflect human input may be compromised by LLM-generated responses. Existing LLM detection approaches often rely on high-dimensional training data such as text, making them unsuitable for structured annotation tasks like multiple-choice labeling. In this work, we investigate the potential of peer prediction --- a mechanism that evaluates the information within workers' responses --- to mitigate LLM-assisted cheating in crowdsourcing with a focus on annotation tasks. Our method quantifies the correlations between worker answers while conditioning on (a subset of) LLM-generated labels available to the requester. Building on prior research, we propose a training-free scoring mechanism with theoretical guarantees under a novel model that accounts for LLM collusion. We establish conditions under which our method is effective and empirically demonstrate its robustness in detecting low-effort cheating on real-world crowdsourcing datasets.

Decentralized federated learning (DFL) uses peer-topeer communication to avoid the single point of failure problem in federated learning and has been considered an attractive solution for machine learning tasks on distributed devices. We provide the first solution to a fundamental network problem of DFL: what overlay network should DFL use to achieve fast training of highly accurate models, low communication, and decentralized construction and maintenance? Overlay topologies of DFL have been investigated, but no existing DFL topology includes decentralized protocols for network construction and topology maintenance. Without these protocols, DFL cannot run in practice. This work presents an overlay network, called FedLay, which provides fast training and low communication cost for practical DFL. FedLay is the first solution for constructing near-random regular topologies in a decentralized manner and maintaining the topologies under node joins and failures. Experiments based on prototype implementation and simulations show that FedLay achieves the fastest model convergence and highest accuracy on real datasets compared to existing DFL solutions while incurring small communication costs and being resilient to node joins and failures.

The pursuit of fairness in machine learning (ML), ensuring that the models do not exhibit biases toward protected demographic groups, typically results in a compromise scenario. This compromise can be explained by a Pareto frontier where given certain resources (e.g., data), reducing the fairness violations often comes at the cost of lowering the model accuracy. In this work, we aim to train models that mitigate group fairness disparity without causing harm to model accuracy. Intuitively, acquiring more data is a natural and promising approach to achieve this goal by reaching a better Pareto frontier of the fairness-accuracy tradeoff. The current data acquisition methods, such as fair active learning approaches, typically require annotating sensitive attributes. However, these sensitive attribute annotations should be protected due to privacy and safety concerns. In this paper, we propose a tractable active data sampling algorithm that does not rely on training group annotations, instead only requiring group annotations on a small validation set. Specifically, the algorithm first scores each new example by its influence on fairness and accuracy evaluated on the validation dataset, and then selects a certain number of examples for training. We theoretically analyze how acquiring more data can improve fairness without causing harm, and validate the possibility of our sampling approach in the context of risk disparity. We also provide the upper bound of generalization error and risk disparity as well as the corresponding connections. Extensive experiments on real-world data demonstrate the effectiveness of our proposed algorithm. Our code is available at [github.com/UCSC-REAL/FairnessWithoutHarm](https://github.com/UCSC-REAL/FairnessWithoutHarm).

The emergence of edge intelligence has made smart IoT services (e.g.,video/audio surveillance, autonomous driving and smart city) a reality. To ensure the quality of service, edge service providers train unbiased models of distributed machine learning jobs over the local datasets collected by edge networks, and usually adopt the parameter server (PS) architecture. However, the training ofunbiased distributed learning(UDL) depends on geo-distributed data and edge resources, bringing a new challenge for service providers: how to effectively schedule and price UDL jobs such that the long-term system utility (i.e.,social welfare) can be maximized. In this paper, we propose an online auction-based scheduling algorithmEris, which determines the data workload, the number and the placement of concurrent workers and PSs for each arriving UDL job, and dynamically prices limited edge resources based on current resource consumption.Erisapplies a primal-dual framework which calls an efficient dual subroutine to schedule UDL jobs, achieving a good competitive ratio and pseudo-polynomial time complexity. To evaluate the effectiveness ofEris, we implement both a testbed and a large-scaled simulator. The results demonstrate thatErisoutperforms and achieves up to 44% more social welfare compared to state-of-the-art algorithms in today's cloud system.

Distributed machine learning (ML) has played a key role in today's proliferation of AI services. A typical model of distributed ML is to partition training datasets over multiple worker nodes to update model parameters in parallel, adopting aparameter serverorAllReducearchitecture. ML training jobs are typically resource elastic, completed using various time lengths with different resource configurations. A fundamental problem in a distributed ML cluster is how to explore the demand elasticity of ML jobs and schedule them with different resource configurations, such that the utilization of resources is maximized and average job completion time is minimized. To address it, we propose an online scheduling algorithm to decide the execution time window, the number and the type of concurrent workers and parameter servers for each job upon its arrival, with a goal of minimizing the weighted average completion time. Our online algorithm consists of (i) an online scheduling framework that groups unprocessed ML training jobs into a batch iteratively, and (ii) a batch scheduling algorithm that configures each ML job to maximize the total weight of scheduled jobs in the current iteration. Our online algorithm guarantees a good parameterized competitive ratio with polynomial time complexity. Extensive evaluations using real-world data demonstrate that it outperforms state-of-the-art schedulers in today's AI cloud systems.

Federated Learning (FL) is a new distributed machine learning (ML) approach which enables thousands of mobile devices to collaboratively train artificial intelligence (AI) models using local data without compromising user privacy. Although FL represents a promising computing paradigm, such training process can not be fully realized without an appropriate economic mechanism that incentivizes the participation of heterogeneous clients. This work targets social cost minimization, and studies the incentive mechanism design in FL through a procurement auction. Different from existing literature, we consider a practical scenario of FL where clients are selected and scheduled at different global iterations to guarantee the completion of the FL job, and capture the distinct feature of FL that the number of global iterations is determined by the local accuracy of all participants to balance between computation and communication. Our auction framework$A_{FL}$first decomposes the social cost minimization problem into a series of winner determination problems (WDPs) based on the number of global iterations. To solve each WDP,$A_{FL}$invokes a greedy algorithm to determine the winners, and a payment algorithm for computing remuneration to winners. Finally,$A_{FL}$returns the best solution among all WDPs. We carried out theoretical analysis to prove that$A_{FL}$is truthful, individual rational, computationally efficient, and achieves a near-optimal social cost. We further extend our model to consider multiple FL jobs with corresponding budgets and propose another efficient algorithm$A_{FL-M}$to solve the extended problem. We conduct large-scale simulations based on the real-world data and testbed experiments by adopting FL frameworks FAVOR and CoCoA. Simulation and experiment results show that both$A_{FL}$and$A_{FL-M}$can reduce the social cost by up to 55% compared with state-of-the-art algorithms.

5G and Internet of Things stimulate smart applications of edge computing, such as autonomous driving and smart city. As edge computing power increases, more and more machine learning (ML) jobs will be trained in the edge-cloud network, adopting the parameter server (PS) architecture. Due to the distinct features of the edge (low-latency and the scarcity of resources), the cloud (high delay and rich computing capacity) and ML jobs (frequent communication between workers and PSs and unfixed runtime), existing cloud job pricing and scheduling algorithms are not applicable. Therefore, how to price, deploy and schedule ML jobs in the edge-cloud network becomes a challenging problem. To solve it, we propose an auction-based online framework DPS. DPS consists of three major parts: job admission control, price function design and scheduling orchestrator. DPS dynamically prices workers and PSs based on historical job information and real-time system status, and decides whether to accept the job according to the deployment cost. DPS then deploys and schedules accepted ML jobs to pursue the maximum social welfare. Through theoretical analysis, we prove that DPS can achieve a good competition ratio and truthfulness in polynomial time. Large-scale simulations and testbed experiments show that DPS can improve social welfare by at least$95\%$, compared with benchmark algorithms in today's cloud system.

Federated Learning (FL) is a new distributed machine learning (ML) approach which enables thousands of mobile devices to collaboratively train artificial intelligence (AI) models using local data without compromising user privacy. Although FL represents a promising computing paradigm, such training process can not be fully realized without an appropriate economic mechanism that incentivizes the participation of heterogeneous clients. This work targets social cost minimization, and studies the incentive mechanism design in FL through a procurement auction. Different from existing literature, we consider a practical scenario of FL where clients are selected and scheduled at different global iterations to guarantee the completion of the FL job, and capture the distinct feature of FL that the number of global iterations is determined by the local accuracy of all participants to balance between computation and communication. Our auction framework$A_{FL}$first decomposes the social cost minimization problem into a series of winner determination problems (WDPs) based on the number of global iterations. Then to solve each WDP,$A_{FL}$invokes a greedy algorithm to determine the winners, and a payment algorithm for computing remuneration to winners. Finally,$A_{FL}$returns the best solution among all WDPs. Theoretical analysis proves that$A_{FL}$is truthful, individual rational, computationally efficient, and achieves a near-optimal social cost. We further conduct large-scale simulation studies based on the real-world data. Simulation results show that$A_{FL}$can reduce the social cost by up to 75% compared with state-of-the-art algorithms.

To realize high quality smart IoT services, such as intelligent video surveillance in Auto Driving and Smart City, tremendous amount of distributed machine learning jobs train unbiased models in wireless edge networks, adopting the parameter server (PS) architecture. Due to the large datasets collected geo-distributedly, the training of unbiased distributed learning (UDL) brings high response latency and bandwidth consumption. In this paper, we propose an online scheduling algorithm, Okita, to minimize both the latency cost and bandwidth cost in UDL. Okita schedules UDL jobs at each time slot to jointly decide the execution time window, the amount of training data, the number and the location of concurrent workers and PSs in each site. To evaluate the practical performance of Okita, we implement a testbed based on Kubernetes. Extensive experiments and simulations show that Okita can reduce up to 60% of total cost, compared with the state-of-the-art schedulers in cloud systems.