Andy Song

Yameng Peng, Andy Song, Vic Ciesielski et al.

Neural architecture search (NAS) aims to automate architecture engineering in neural networks. This often requires a high computational overhead to evaluate a number of candidate networks from the set of all possible networks in the search space during the search. Prediction of the networks' performance can alleviate this high computational overhead by mitigating the need for evaluating every candidate network. Developing such a predictor typically requires a large number of evaluated architectures which may be difficult to obtain. We address this challenge by proposing a novel evolutionary-based NAS strategy, Predictor-assisted E-NAS (PRE-NAS), which can perform well even with an extremely small number of evaluated architectures. PRE-NAS leverages new evolutionary search strategies and integrates high-fidelity weight inheritance over generations. Unlike one-shot strategies, which may suffer from bias in the evaluation due to weight sharing, offspring candidates in PRE-NAS are topologically homogeneous, which circumvents bias and leads to more accurate predictions. Extensive experiments on NAS-Bench-201 and DARTS search spaces show that PRE-NAS can outperform state-of-the-art NAS methods. With only a single GPU searching for 0.6 days, competitive architecture can be found by PRE-NAS which achieves 2.40% and 24% test error rates on CIFAR-10 and ImageNet respectively.

Danru Xu, Erdun Gao, Wei Huang et al.

Learning the underlying Bayesian Networks (BNs), represented by directed acyclic graphs (DAGs), of the concerned events from purely-observational data is a crucial part of evidential reasoning. This task remains challenging due to the large and discrete search space. A recent flurry of developments followed NOTEARS[1] recast this combinatorial problem into a continuous optimization problem by leveraging an algebraic equality characterization of acyclicity. However, the continuous optimization methods suffer from obtaining non-spare graphs after the numerical optimization, which leads to the inflexibility to rule out the potentially cycle-inducing edges or false discovery edges with small values. To address this issue, in this paper, we develop a completely data-driven DAG structure learning method without a predefined value to post-threshold small values. We name our method NOTEARS with adaptive Lasso (NOTEARS-AL), which is achieved by applying the adaptive penalty method to ensure the sparsity of the estimated DAG. Moreover, we show that NOTEARS-AL also inherits the oracle properties under some specific conditions. Extensive experiments on both synthetic and a real-world dataset demonstrate that our method consistently outperforms NOTEARS.

Yameng Peng, Andy Song, HaythamM. Fayek et al.

Zero-shot proxies, also known as training-free metrics, are widely adopted to reduce the computational overhead in neural network evaluation for scenarios such as Neural Architecture Search (NAS), as they do not require any training. Existing zero-shot metrics have several limitations, including weak correlation with the true performance and poor generalisation across different networks or downstream tasks. For example, most of these metrics apply only to either convolutional neural networks (CNNs) or Transformers, but not both. To address these limitations, we propose Sample-Wise Activation Patterns (SWAP), and its derivative, SWAP-Score, a novel and highly effective zero-shot metric. SWAP-Score is broadly applicable across both architecture families and task domains, demonstrating strong predictive performance in the majority of tasks. This metric measures the expressivity of neural networks over a mini-batch of samples, showing a high correlation with the neural networks' ground-truth performance. For both CNNs and Transformers, the SWAP-Score outperforms existing zero-shot metrics across computer vision and natural language processing tasks. For instance, Spearman's correlation coefficient between the SWAP-Score and CIFAR-10 validation accuracy for DARTS CNNs is 0.93, and 0.71 for FlexiBERT Transformers on GLUE tasks. Moreover, SWAP-Score is label-independent, hence can be applied at the pre-training stage of language models to estimate their performance for downstream tasks. When applied to NAS, SWAP-empowered NAS, SWAP-NAS can achieve competitive performance using only approximately 6 and 9 minutes of GPU time, on CIFAR-10 and ImageNet respectively. Our code is available at: https://github.com/pym1024/SWAP_Universal

Sihao Lin, Pumeng Lyu, Dongrui Liu et al.

Self-attention mechanism is the key of the Transformer but often criticized for its computation demands. Previous token pruning works motivate their methods from the view of computation redundancy but still need to load the full network and require same memory costs. This paper introduces a novel strategy that simplifies vision transformers and reduces computational load through the selective removal of non-essential attention layers, guided by entropy considerations. We identify that regarding the attention layer in bottom blocks, their subsequent MLP layers, i.e. two feed-forward layers, can elicit the same entropy quantity. Meanwhile, the accompanied MLPs are under-exploited since they exhibit smaller feature entropy compared to those MLPs in the top blocks. Therefore, we propose to integrate the uninformative attention layers into their subsequent counterparts by degenerating them into identical mapping, yielding only MLP in certain transformer blocks. Experimental results on ImageNet-1k show that the proposed method can remove 40% attention layer of DeiT-B, improving throughput and memory bound without performance compromise. Code is available at https://github.com/sihaoevery/lambda_vit.

Long Chen, Xiaotian Song, Andy Song et al.

Spiking Large Language Models have been shown as a good alternative to LLMs in various scenarios. Existing methods for creating Spiking LLMs, i.e., direct training and ANN-SNN conversion, often suffer from performance degradation and relatively high computational costs. To address these issues, we propose a novel Fast ANN-SNN conversion strategy (FAS) that transforms LLMs into spiking LLMs in two stages. The first stage employs a full-parameter fine-tuning of pre-trained models, so it does not need any direct training from scratch. The second stage introduces a coarse-to-fine calibration method to reduce conversion errors and improve accuracy. Experiments on both language and vision-language tasks across four different scales of LLMs demonstrate that FAS can achieve state-of-the-art performance yet with significantly reduced inference latency and computational costs. Notably, FAS only takes eight timesteps to achieve an accuracy of 3\% higher than that of the OPT-7B model, while reducing energy consumption by 96.63\%. The source code is available at https://github.com/lc783/FAS

Yameng Peng, Andy Song, Haytham M. Fayek et al.

Training-free metrics (a.k.a. zero-cost proxies) are widely used to avoid resource-intensive neural network training, especially in Neural Architecture Search (NAS). Recent studies show that existing training-free metrics have several limitations, such as limited correlation and poor generalisation across different search spaces and tasks. Hence, we propose Sample-Wise Activation Patterns and its derivative, SWAP-Score, a novel high-performance training-free metric. It measures the expressivity of networks over a batch of input samples. The SWAP-Score is strongly correlated with ground-truth performance across various search spaces and tasks, outperforming 15 existing training-free metrics on NAS-Bench-101/201/301 and TransNAS-Bench-101. The SWAP-Score can be further enhanced by regularisation, which leads to even higher correlations in cell-based search space and enables model size control during the search. For example, Spearman's rank correlation coefficient between regularised SWAP-Score and CIFAR-100 validation accuracies on NAS-Bench-201 networks is 0.90, significantly higher than 0.80 from the second-best metric, NWOT. When integrated with an evolutionary algorithm for NAS, our SWAP-NAS achieves competitive performance on CIFAR-10 and ImageNet in approximately 6 minutes and 9 minutes of GPU time respectively.

Muxing Li, Zesheng Ye, Sharon Li et al.

To detect unauthorized data usage in training large-scale generative models (e.g., ChatGPT or Midjourney), membership inference attacks (MIA) have proven effective in distinguishing a single training instance (a member) from a single non-training instance (a non-member). This success is mainly credited to a memorization effect: models tend to perform better on a member than a non-member. However, we find that standard MIAs fail against distilled generative models (i.e., student models) that are increasingly deployed in practice for efficiency (e.g., ChatGPT 4o-mini). Trained exclusively on data generated from a large-scale model (a teacher model), the student model lacks direct exposure to any members (teacher's training data), nullifying the memorization effect that standard MIAs rely on. This finding reveals a serious privacy loophole, where generation-service providers could deploy a student model whose teacher was potentially trained on unauthorized data, yet claim the deployed model is clean because it was not directly trained on such data. Hence, are distilled models inherently unauditable for upstream privacy violations, and should we discard them when we care about privacy? We contend no, as we uncover a memory chain connecting the student and teacher's member data: the distribution of student-generated data aligns more closely with the distribution of the teacher's members than with non-members, thus we can detect unauthorized data usage even when direct instance-level memorization is absent. This leads us to posit that MIAs on distilled generative models should shift from instance-level scores to distribution-level statistics. We further propose three principles of distribution-based MIAs for detecting unauthorized training data through distilled generative models, and validate our position through an exemplar framework. We lastly discuss the implications of our position.

Sihao Lin, Daqi Liu, Ruochong Fu et al.

Estimating the 3D world from 2D monocular images is a fundamental yet challenging task due to the labour-intensive nature of 3D annotations. To simplify label acquisition, this work proposes a novel approach that bridges 2D vision foundation models (VFMs) with 3D tasks by decoupling 3D supervision into an ensemble of image-level primitives, e.g., semantic and geometric components. As a key motivator, we leverage the zero-shot capabilities of vision-language models for image semantics. However, due to the notorious ill-posed problem - multiple distinct 3D scenes can produce identical 2D projections, directly inferring metric depth from a monocular image in a zero-shot manner is unsuitable. In contrast, 2D VFMs provide promising sources of relative depth, which theoretically aligns with metric depth when properly scaled and offset. Thus, we adapt the relative depth derived from VFMs into metric depth by optimising the scale and offset using temporal consistency, also known as novel view synthesis, without access to ground-truth metric depth. Consequently, we project the semantics into 3D space using the reconstructed metric depth, thereby providing 3D supervision. Extensive experiments on nuScenes and SemanticKITTI demonstrate the effectiveness of our framework. For instance, the proposed method surpasses the current state-of-the-art by 3.34% mIoU on nuScenes for voxel occupancy prediction.

Xiangtan Lin, Pengzhen Ren, Chung-Hsing Yeh et al.

Person re-identification (Re-ID) has been a significant research topic in the past decade due to its real-world applications and research significance. While supervised person Re-ID methods achieve superior performance over unsupervised counterparts, they can not scale to large unlabelled datasets and new domains due to the prohibitive labelling cost. Therefore, unsupervised person Re-ID has drawn increasing attention for its potential to address the scalability issue in person Re-ID. Unsupervised person Re-ID is challenging primarily due to lacking identity labels to supervise person feature learning. The corresponding solutions are diverse and complex, with various merits and limitations. Therefore, comprehensive surveys on this topic are essential to summarise challenges and solutions to foster future research. Existing person Re-ID surveys have focused on supervised methods from classifications and applications but lack detailed discussion on how the person Re-ID solutions address the underlying challenges. This survey review recent works on unsupervised person Re-ID from the perspective of challenges and solutions. Specifically, we provide an in-depth analysis of highly influential methods considering the four significant challenges in unsupervised person Re-ID: 1) lacking ground-truth identity labels to supervise person feature learning; 2) learning discriminative person features with pseudo-supervision; 3) learning cross-camera invariant person feature, and 4) the domain shift between datasets. We summarise and analyse evaluation results and provide insights on the effectiveness of the solutions. Finally, we discuss open issues and suggest some promising future research directions.

Mohammad Saiedur Rahaman, Wei Shao, Flora D. Salim et al.

Existing parking recommendation solutions mainly focus on finding and suggesting parking spaces based on the unoccupied options only. However, there are other factors associated with parking spaces that can influence someone's choice of parking such as fare, parking rule, walking distance to destination, travel time, likelihood to be unoccupied at a given time. More importantly, these factors may change over time and conflict with each other which makes the recommendations produced by current parking recommender systems ineffective. In this paper, we propose a novel problem called multi-objective parking recommendation. We present a solution by designing a multi-objective parking recommendation engine called MoParkeR that considers various conflicting factors together. Specifically, we utilise a non-dominated sorting technique to calculate a set of Pareto-optimal solutions, consisting of recommended trade-off parking spots. We conduct extensive experiments using two real-world datasets to show the applicability of our multi-objective recommendation methodology.

Wei Shao, Sichen Zhao, Zhen Zhang et al.

Existing research on parking availability sensing mainly relies on extensive contextual and historical information. In practice, the availability of such information is a challenge as it requires continuous collection of sensory signals. In this study, we design an end-to-end transfer learning framework for parking availability sensing to predict parking occupancy in areas in which the parking data is insufficient to feed into data-hungry models. This framework overcomes two main challenges: 1) many real-world cases cannot provide enough data for most existing data-driven models, and 2) it is difficult to merge sensor data and heterogeneous contextual information due to the differing urban fabric and spatial characteristics. Our work adopts a widely-used concept, adversarial domain adaptation, to predict the parking occupancy in an area without abundant sensor data by leveraging data from other areas with similar features. In this paper, we utilise more than 35 million parking data records from sensors placed in two different cities, one a city centre and the other a coastal tourist town. We also utilise heterogeneous spatio-temporal contextual information from external resources, including weather and points of interest. We quantify the strength of our proposed framework in different cases and compare it to the existing data-driven approaches. The results show that the proposed framework is comparable to existing state-of-the-art methods and also provide some valuable insights on parking availability prediction.

Yong Liu, Lin Shang, Andy Song

Typical person re-identification (re-ID) methods train a deep CNN to extract deep features and combine them with a distance metric for the final evaluation. In this work, we focus on exploiting the full information encoded in the deep feature to boost the re-ID performance. First, we propose a Deep Feature Fusion (DFF) method to exploit the diverse information embedded in a deep feature. DFF treats each sub-feature as an information carrier and employs a diffusion process to exchange their information. Second, we propose an Adaptive Re-Ranking (ARR) method to exploit the contextual information encoded in the features of neighbors. ARR utilizes the contextual information to re-rank the retrieval results in an iterative manner. Particularly, it adds more contextual information after each iteration automatically to consider more matches. Third, we propose a strategy that combines DFF and ARR to enhance the performance. Extensive comparative evaluations demonstrate the superiority of the proposed methods on three large benchmarks.

Shivanthan Yohanandan, Andy Song, Adrian G. Dyer et al.

Retraction due to significant oversight

Min Chen, Andy Song, Shivanthan A. C. Yhanandan et al.

Image processing and recognition are an important part of the modern society, with applications in fields such as advanced artificial intelligence, smart assistants, and security surveillance. The essential first step involved in almost all the visual tasks is background subtraction with a static camera. Ensuring that this critical step is performed in the most efficient manner would therefore improve all aspects related to objects recognition and tracking, behavior comprehension, etc.. Although background subtraction method has been applied for many years, its application suffers from real-time requirement. In this letter, we present a novel approach in implementing the background subtraction. The proposed method uses compressed, low-resolution grayscale image for the background subtraction. These low-resolution grayscale images were found to preserve the salient information very well. To verify the feasibility of our methodology, two prevalent methods, ViBe and GMM, are used in the experiment. The results of the proposed methodology confirm the effectiveness of our approach.

Shivanthan A. C. Yohanandan, Adrian G. Dyer, Dacheng Tao et al.

Visual salience detection originated over 500 million years ago and is one of nature's most efficient mechanisms. In contrast, many state-of-the-art computational saliency models are complex and inefficient. Most saliency models process high-resolution color (HC) images; however, insights into the evolutionary origins of visual salience detection suggest that achromatic low-resolution vision is essential to its speed and efficiency. Previous studies showed that low-resolution color and high-resolution grayscale images preserve saliency information. However, to our knowledge, no one has investigated whether saliency is preserved in low-resolution grayscale (LG) images. In this study, we explain the biological and computational motivation for LG, and show, through a range of human eye-tracking and computational modeling experiments, that saliency information is preserved in LG images. Moreover, we show that using LG images leads to significant speedups in model training and detection times and conclude by proposing LG images for fast and efficient salience detection.