Yiwei Chen

Justin Lovelace, Varsha Kishore, Yiwei Chen et al. · cmu

Current language models demonstrate remarkable proficiency in text generation. However, for many applications it is desirable to control attributes, such as sentiment, or toxicity, of the generated language -- ideally tailored towards each specific use case and target audience. For auto-regressive language models, existing guidance methods are prone to decoding errors that cascade during generation and degrade performance. In contrast, text diffusion models can easily be guided with, for example, a simple linear sentiment classifier -- however they do suffer from significantly higher perplexity than auto-regressive alternatives. In this paper we use a guided diffusion model to produce a latent proposal that steers an auto-regressive language model to generate text with desired properties. Our model inherits the unmatched fluency of the auto-regressive approach and the plug-and-play flexibility of diffusion. We show that it outperforms previous plug-and-play guidance methods across a wide range of benchmark data sets. Further, controlling a new attribute in our framework is reduced to training a single logistic regression classifier.

Zhihao Zhang, Yiwei Chen, Weizhan Zhang et al.

Viewport prediction is a crucial aspect of tile-based 360 video streaming system. However, existing trajectory based methods lack of robustness, also oversimplify the process of information construction and fusion between different modality inputs, leading to the error accumulation problem. In this paper, we propose a tile classification based viewport prediction method with Multi-modal Fusion Transformer, namely MFTR. Specifically, MFTR utilizes transformer-based networks to extract the long-range dependencies within each modality, then mine intra- and inter-modality relations to capture the combined impact of user historical inputs and video contents on future viewport selection. In addition, MFTR categorizes future tiles into two categories: user interested or not, and selects future viewport as the region that contains most user interested tiles. Comparing with predicting head trajectories, choosing future viewport based on tile's binary classification results exhibits better robustness and interpretability. To evaluate our proposed MFTR, we conduct extensive experiments on two widely used PVS-HM and Xu-Gaze dataset. MFTR shows superior performance over state-of-the-art methods in terms of average prediction accuracy and overlap ratio, also presents competitive computation efficiency.

Gongxin Yao, Yixin Xuan, Yiwei Chen et al.

Image-to-point cloud registration aims to determine the relative camera pose between an RGB image and a reference point cloud, serving as a general solution for locating 3D objects from 2D observations. Matching individual points with pixels can be inherently ambiguous due to modality gaps. To address this challenge, we propose a framework to capture quantity-aware correspondences between local point sets and pixel patches and refine the results at both the point and pixel levels. This framework aligns the high-level semantics of point sets and pixel patches to improve the matching accuracy. On a coarse scale, the set-to-patch correspondence is expected to be influenced by the quantity of 3D points. To achieve this, a novel supervision strategy is proposed to adaptively quantify the degrees of correlation as continuous values. On a finer scale, point-to-pixel correspondences are refined from a smaller search space through a well-designed scheme, which incorporates both resampling and quantity-aware priors. Particularly, a confidence sorting strategy is proposed to proportionally select better correspondences at the final stage. Leveraging the advantages of high-quality correspondences, the problem is successfully resolved using an efficient Perspective-n-Point solver within the framework of random sample consensus (RANSAC). Extensive experiments on the KITTI Odometry and NuScenes datasets demonstrate the superiority of our method over the state-of-the-art methods.

Ziqiang Zheng, Yiwei Chen, Jipeng Zhang et al.

Large language models (LLMs) have demonstrated a powerful ability to answer various queries as a general-purpose assistant. The continuous multi-modal large language models (MLLM) empower LLMs with the ability to perceive visual signals. The launch of GPT-4 (Generative Pre-trained Transformers) has generated significant interest in the research communities. GPT-4V(ison) has demonstrated significant power in both academia and industry fields, as a focal point in a new artificial intelligence generation. Though significant success was achieved by GPT-4V, exploring MLLMs in domain-specific analysis (e.g., marine analysis) that required domain-specific knowledge and expertise has gained less attention. In this study, we carry out the preliminary and comprehensive case study of utilizing GPT-4V for marine analysis. This report conducts a systematic evaluation of existing GPT-4V, assessing the performance of GPT-4V on marine research and also setting a new standard for future developments in MLLMs. The experimental results of GPT-4V show that the responses generated by GPT-4V are still far away from satisfying the domain-specific requirements of the marine professions. All images and prompts used in this study will be available at https://github.com/hkust-vgd/Marine_GPT-4V_Eval

Yiwei Chen, Chen Jiang, Yu Pan

Single-Photon Image Super-Resolution (SPISR) aims to recover a high-resolution volumetric photon counting cube from a noisy low-resolution one by computational imaging algorithms. In real-world scenarios, pairs of training samples are often expensive or impossible to obtain. By extending Equivariant Imaging (EI) to volumetric single-photon data, we propose a self-supervised learning framework for the SPISR task. Particularly, using the Poisson unbiased Kullback-Leibler risk estimator and equivariance, our method is able to learn from noisy measurements without ground truths. Comprehensive experiments on simulated and real-world dataset demonstrate that the proposed method achieves comparable performance with supervised learning and outperforms interpolation-based methods.

Yiwei Chen, Kehuan Yan, Yu Pan et al.

Quantum theory provides non-classical principles, such as superposition and entanglement, that inspires promising paradigms in machine learning. However, most existing quantum-inspired fusion models rely solely on unitary or unitary-like transformations to generate quantum entanglement. While theoretically expressive, such approaches often suffer from training instability and limited generalizability. In this work, we propose a Quantum-inspired Neural Network with Quantum Jump (QiNN-QJ) for multimodal entanglement modelling. Each modality is firstly encoded as a quantum pure state, after which a differentiable module simulating the QJ operator transforms the separable product state into the entangled representation. By jointly learning Hamiltonian and Lindblad operators, QiNN-QJ generates controllable cross-modal entanglement among modalities with dissipative dynamics, where structured stochasticity and steady-state attractor properties serve to stabilize training and constrain entanglement shaping. The resulting entangled states are projected onto trainable measurement vectors to produce predictions. In addition to achieving superior performance over the state-of-the-art models on benchmark datasets, including CMU-MOSI, CMU-MOSEI, and CH-SIMS, QiNN-QJ facilitates enhanced post-hoc interpretability through von-Neumann entanglement entropy. This work establishes a principled framework for entangled multimodal fusion and paves the way for quantum-inspired approaches in modelling complex cross-modal correlations.

Yuk-Kwan Wong, Haixin Liang, Zeyu Ma et al.

Marine visual understanding is essential for monitoring and protecting marine ecosystems, enabling automatic and scalable biological surveys. However, progress is hindered by limited training data and the lack of a systematic task formulation that aligns domain-specific marine challenges with well-defined computer vision tasks, thereby limiting effective model application. To address this gap, we present ORCA, a multi-modal benchmark for marine research comprising 14,647 images from 478 species, with 42,217 bounding box annotations and 22,321 expert-verified instance captions. The dataset provides fine-grained visual and textual annotations that capture morphology-oriented attributes across diverse marine species. To catalyze methodological advances, we evaluate 18 state-of-the-art models on three tasks: object detection (closed-set and open-vocabulary), instance captioning, and visual grounding. Results highlight key challenges, including species diversity, morphological overlap, and specialized domain demands, underscoring the difficulty of marine understanding. ORCA thus establishes a comprehensive benchmark to advance research in marine domain. Project Page: http://orca.hkustvgd.com/.

Hadi Reisizadeh, Jiajun Ruan, Yiwei Chen et al.

Unlearning in large language models (LLMs) is critical for regulatory compliance and for building ethical generative AI systems that avoid producing private, toxic, illegal, or copyrighted content. Despite rapid progress, in this work we show that \textit{almost all} existing unlearning methods fail to achieve true forgetting in practice. Specifically, while evaluations of these `unlearned' models under deterministic (greedy) decoding often suggest successful knowledge removal using standard benchmarks (as has been done in the literature), we show that sensitive information reliably resurfaces when models are sampled with standard probabilistic decoding. To rigorously capture this vulnerability, we introduce \texttt{leak@$k$}, a new meta-evaluation metric that quantifies the likelihood of forgotten knowledge reappearing when generating $k$ samples from the model under realistic decoding strategies. Using three widely adopted benchmarks, TOFU, MUSE, and WMDP, we conduct the first large-scale, systematic study of unlearning reliability using our newly defined \texttt{leak@$k$} metric. Our findings demonstrate that knowledge leakage persists across methods and tasks, underscoring that current state-of-the-art unlearning techniques provide only limited forgetting and highlighting the urgent need for more robust approaches to LLM unlearning.

Bingqi Shang, Yiwei Chen, Yihua Zhang et al.

Large language model (LLM) unlearning has become a critical mechanism for removing undesired data, knowledge, or behaviors from pre-trained models while retaining their general utility. Yet, with the rise of open-weight LLMs, we ask: can the unlearning process itself be backdoored, appearing successful under normal conditions yet reverting to pre-unlearned behavior when a hidden trigger is activated? Drawing inspiration from classical backdoor attacks that embed triggers into training data to enforce specific behaviors, we investigate backdoor unlearning, where models forget as intended in the clean setting but recover forgotten knowledge when the trigger appears. We show that designing such attacks presents unique challenges, hinging on where triggers are placed and how backdoor training is reinforced. We uncover a strong link between backdoor efficacy and the attention sink phenomenon, i.e., shallow input tokens consistently attract disproportionate attention in LLMs. Our analysis reveals that these attention sinks serve as gateways for backdoor unlearning: placing triggers at sink positions and aligning their attention values markedly enhances backdoor persistence. Extensive experiments validate these findings, showing that attention-sink-guided backdoor unlearning reliably restores forgotten knowledge in the presence of backdoor triggers, while behaving indistinguishably from a normally unlearned model when triggers are absent. Code is available at https://github.com/OPTML-Group/Unlearn-Backdoor.

Yiwei Chen, Jingwen Qian, Junming Wang et al.

We introduce the Online Preference Reporting and Aggregation (OPRA) system, an open-source online system that aims at providing support for group decision-making. We illustrate OPRA's distinctive features: UI for reporting rankings with ties, comprehensive analytics of preferences, and group decision-making in combinatorial domains. We also discuss our work in an automatic mentor matching system. We hope that the open-source nature of OPRA will foster the development of computerized group decision support systems.

Yiwei Chen, Yuguang Yao, Yihua Zhang et al.

Recent vision language models (VLMs) have made remarkable strides in generative modeling with multimodal inputs, particularly text and images. However, their susceptibility to generating harmful content when exposed to unsafe queries raises critical safety concerns. While current alignment strategies primarily rely on supervised safety fine-tuning with curated datasets, we identify a fundamental limitation we call the ''safety mirage'', where supervised fine-tuning inadvertently reinforces spurious correlations between superficial textual patterns and safety responses, rather than fostering deep, intrinsic mitigation of harm. We show that these spurious correlations leave fine-tuned VLMs vulnerable even to a simple one-word modification-based attack, where substituting a single word in text queries with a spurious correlation-inducing alternative can effectively bypass safeguards. Additionally, these correlations contribute to the over-prudence, causing fine-tuned VLMs to refuse benign queries unnecessarily. To address these issues, we show machine unlearning (MU) as a powerful alternative to supervised safety fine-tuning, as it avoids biased feature-label mappings and directly removes harmful knowledge from VLMs while preserving their general capabilities. Extensive evaluations across safety benchmarks show that under MU-based alignment reduces the attack success rate by up to 60.17% and cuts unnecessary rejections by over 84.20%. WARNING: There exist AI generations that may be offensive in nature.

Yiwei Chen, Soumyadeep Pal, Yimeng Zhang et al.

Machine unlearning (MU) for large language models (LLMs), commonly referred to as LLM unlearning, seeks to remove specific undesirable data or knowledge from a trained model, while maintaining its performance on standard tasks. While unlearning plays a vital role in protecting data privacy, enforcing copyright, and mitigating sociotechnical harms in LLMs, we identify a new vulnerability post-unlearning: unlearning trace detection. We discover that unlearning leaves behind persistent ''fingerprints'' in LLMs, detectable traces in both model behavior and internal representations. These traces can be identified from output responses, even when prompted with forget-irrelevant inputs. Specifically, even a simple supervised classifier can determine whether a model has undergone unlearning, using only its prediction logits or even its textual outputs. Further analysis shows that these traces are embedded in intermediate activations and propagate nonlinearly to the final layer, forming low-dimensional, learnable manifolds in activation space. Through extensive experiments, we demonstrate that unlearning traces can be detected with over 90% accuracy even under forget-irrelevant inputs, and that larger LLMs exhibit stronger detectability. These findings reveal that unlearning leaves measurable signatures, introducing a new risk of reverse-engineering forgotten information when a model is identified as unlearned, given an input query.

Zizhao Chen, Mustafa Omer Gul, Yiwei Chen et al.

Multi-turn interactions between large language models (LLMs) and users naturally include implicit feedback signals. If an LLM responds in an unexpected way to an instruction, the user is likely to signal it by rephrasing the request, expressing frustration, or pivoting to an alternative task. Such signals are task-independent and occupy a relatively constrained subspace of language, allowing the LLM to identify them even if it fails on the actual task. We introduce ReSpect, a method to learn from such signals in past interactions via retrospection without additional annotations. We deploy ReSpect in a new multimodal interaction scenario, where humans instruct a multimodal LLM to solve an abstract reasoning task with a combinatorial solution space. Through thousands of interactions with humans, we show how ReSpect gradually improves task completion rate from 31% to 82%, all without any external annotation.

Yiwei Chen, Chao Tang, Amir Aghabiglou et al.

We propose a new approach for non-Cartesian magnetic resonance image reconstruction. While unrolled architectures provide robustness via data-consistency layers, embedding measurement operators in Deep Neural Network (DNN) can become impractical at large scale. Alternative Plug-and-Play (PnP) approaches, where the denoising DNNs are blind to the measurement setting, are not affected by this limitation and have also proven effective, but their highly iterative nature also affects scalability. To address this scalability challenge, we leverage the "Residual-to-Residual DNN series for high-Dynamic range imaging (R2D2)" approach recently introduced in astronomical imaging. R2D2's reconstruction is formed as a series of residual images, iteratively estimated as outputs of DNNs taking the previous iteration's image estimate and associated data residual as inputs. The method can be interpreted as a learned version of the Matching Pursuit algorithm. We demonstrate R2D2 in simulation, considering radial k-space sampling acquisition sequences. Our preliminary results suggest that R2D2 achieves: (i) suboptimal performance compared to its unrolled incarnation R2D2-Net, which is however non-scalable due to the necessary embedding of NUFFT-based data-consistency layers; (ii) superior reconstruction quality to a scalable version of R2D2-Net embedding an FFT-based approximation for data consistency; (iii) superior reconstruction quality to PnP, while only requiring few iterations.

Sumanth Varambally, Marshall Fisher, Jas Thakker et al.

Foundation models for weather science are pre-trained on vast amounts of structured numerical data and outperform traditional weather forecasting systems. However, these models lack language-based reasoning capabilities, limiting their utility in interactive scientific workflows. Large language models (LLMs) excel at understanding and generating text but cannot reason about high-dimensional meteorological datasets. We bridge this gap by building a novel agentic framework for weather science. Our framework includes a Python code-based environment for agents (ZephyrusWorld) to interact with weather data, featuring tools like an interface to WeatherBench 2 dataset, geoquerying for geographical masks from natural language, weather forecasting, and climate simulation capabilities. We design Zephyrus, a multi-turn LLM-based weather agent that iteratively analyzes weather datasets, observes results, and refines its approach through conversational feedback loops. We accompany the agent with a new benchmark, ZephyrusBench, with a scalable data generation pipeline that constructs diverse question-answer pairs across weather-related tasks, from basic lookups to advanced forecasting, extreme event detection, and counterfactual reasoning. Experiments on this benchmark demonstrate the strong performance of Zephyrus agents over text-only baselines, outperforming them by up to 35 percentage points in correctness. However, on harder tasks, Zephyrus performs similarly to text-only baselines, highlighting the challenging nature of our benchmark and suggesting promising directions for future work.

Lijie Yang, Zhihao Zhang, Arti Jain et al.

Large reasoning models achieve strong performance through test-time scaling but incur substantial computational overhead, particularly from excessive token generation when processing short input prompts. While sparse attention mechanisms can reduce latency and memory usage, existing approaches suffer from significant accuracy degradation due to accumulated errors during long-generation reasoning. These methods generally require either high token retention rates or expensive retraining. We introduce LessIsMore, a training-free sparse attention mechanism for reasoning tasks, which leverages global attention patterns rather than relying on traditional head-specific local optimizations. LessIsMore aggregates token selections from local attention heads with recent contextual information, enabling unified cross-head token ranking for future decoding layers. This unified selection improves generalization and efficiency by avoiding the need to maintain separate token subsets per head. Evaluation across diverse reasoning tasks and benchmarks shows that LessIsMore preserves -- and in some cases improves -- accuracy while achieving a $1.1\times$ average decoding speed-up compared to full attention. Moreover, LessIsMore attends to $2\times$ fewer tokens without accuracy loss, achieving a $1.13\times$ end-to-end speed-up compared to existing sparse attention methods.

Lu Xu, Jiaqian Yu, Xiongfeng Peng et al.

To meet the growing demand for smarter, faster, and more efficient embodied AI solutions, we introduce a novel Mixture-of-Expert (MoE) method that significantly boosts reasoning and learning efficiency for embodied autonomous systems. General MoE models demand extensive training data and complex optimization, which limits their applicability in embodied AI such as autonomous driving (AD) and robotic manipulation. In this work, we propose a skill-oriented MoE called MoSE, which mimics the human learning and reasoning process skill-by-skill, step-by-step. We introduce a skill-oriented routing mechanism that begins with defining and annotating specific skills, enabling experts to identify the necessary competencies for various scenarios and reasoning tasks, thereby facilitating skill-by-skill learning. To better align with multi-step planning in human reasoning and in end-to-end driving models, we build a hierarchical skill dataset and pretrain the router to encourage the model to think step-by-step. Unlike other multi-round dialogues, MoSE integrates valuable auxiliary tasks (e.g. perception-prediction-planning for AD, and high-level and low-level planning for robots) in one single forward process without introducing any extra computational cost. With less than 3B sparsely activated parameters, our model effectively grows more diverse expertise and outperforms models on both AD corner-case reasoning tasks and robot reasoning tasks with less than 40% of the parameters.

Yiwei Chen, Amir Aghabiglou, Shijie Chen et al.

We introduce the R2D2 Deep Neural Network (DNN) series paradigm for fast and scalable image reconstruction from highly-accelerated non-Cartesian k-space acquisitions in Magnetic Resonance Imaging (MRI). While unrolled DNN architectures provide a robust image formation approach via data-consistency layers, embedding non-uniform fast Fourier transform operators in a DNN can become impractical to train at large scale, e.g in 2D MRI with a large number of coils, or for higher-dimensional imaging. Plug-and-play approaches that alternate a learned denoiser blind to the measurement setting with a data-consistency step are not affected by this limitation but their highly iterative nature implies slow reconstruction. To address this scalability challenge, we leverage the R2D2 paradigm that was recently introduced to enable ultra-fast reconstruction for large-scale Fourier imaging in radio astronomy. R2D2's reconstruction is formed as a series of residual images iteratively estimated as outputs of DNN modules taking the previous iteration's data residual as input. The method can be interpreted as a learned version of the Matching Pursuit algorithm. A series of R2D2 DNN modules were sequentially trained in a supervised manner on the fastMRI dataset and validated for 2D multi-coil MRI in simulation and on real data, targeting highly under-sampled radial k-space sampling. Results suggest that a series with only few DNNs achieves superior reconstruction quality over its unrolled incarnation R2D2-Net (whose training is also much less scalable), and over the state-of-the-art diffusion-based "Decomposed Diffusion Sampler" approach (also characterised by a slower reconstruction process).

Yiwei Chen, Gongxin Yao, Yong Liu et al.

Photon-efficient imaging with the single-photon light detection and ranging (LiDAR) captures the three-dimensional (3D) structure of a scene by only a few detected signal photons per pixel. However, the existing computational methods for photon-efficient imaging are pre-tuned on a restricted scenario or trained on simulated datasets. When applied to realistic scenarios whose signal-to-background ratios (SBR) and other hardware-specific properties differ from those of the original task, the model performance often significantly deteriorates. In this paper, we present a domain adversarial adaptation design to alleviate this domain shift problem by exploiting unlabeled real-world data, with significant resource savings. This method demonstrates superior performance on simulated and real-world experiments using our home-built up-conversion single-photon imaging system, which provides an efficient approach to bypass the lack of ground-truth depth information in implementing computational imaging algorithms for realistic applications.

Gongxin Yao, Yiwei Chen, Yong Liu et al.

Single-photon light detection and ranging (LiDAR) has been widely applied to 3D imaging in challenging scenarios. However, limited signal photon counts and high noises in the collected data have posed great challenges for predicting the depth image precisely. In this paper, we propose a pixel-wise residual shrinkage network for photon-efficient imaging from high-noise data, which adaptively generates the optimal thresholds for each pixel and denoises the intermediate features by soft thresholding. Besides, redefining the optimization target as pixel-wise classification provides a sharp advantage in producing confident and accurate depth estimation when compared with existing research. Comprehensive experiments conducted on both simulated and real-world datasets demonstrate that the proposed model outperforms the state-of-the-arts and maintains robust imaging performance under different signal-to-noise ratios including the extreme case of 1:100.

Yiwei Chen, Yu Pan, Daoyi Dong

States of quantum many-body systems are defined in a high-dimensional Hilbert space, where rich and complex interactions among subsystems can be modelled. In machine learning, complex multiple multilinear correlations may also exist within input features. In this paper, we present a quantum-inspired multilinear model, named Residual Tensor Train (ResTT), to capture the multiple multilinear correlations of features, from low to high orders, within a single model. ResTT is able to build a robust decision boundary in a high-dimensional space for solving fitting and classification tasks. In particular, we prove that the fully-connected layer and the Volterra series can be taken as special cases of ResTT. Furthermore, we derive the rule for weight initialization that stabilizes the training of ResTT based on a mean-field analysis. We prove that such a rule is much more relaxed than that of TT, which means ResTT can easily address the vanishing and exploding gradient problem that exists in the existing TT models. Numerical experiments demonstrate that ResTT outperforms the state-of-the-art tensor network and benchmark deep learning models on MNIST and Fashion-MNIST datasets. Moreover, ResTT achieves better performance than other statistical methods on two practical examples with limited data which are known to have complex feature interactions.

Yiwei Chen, Yu Pan, Guofeng Zhang et al.

Quantum properties, such as entanglement and coherence, are indispensable resources in various quantum information processing tasks. However, there still lacks an efficient and scalable way to detecting these useful features, especially for high-dimensional and multipartite quantum systems. In this work, we exploit the convexity of samples without the desired quantum features and design an unsupervised machine learning method to detect the presence of such features as anomalies. Particularly, in the context of entanglement detection, we propose a complex-valued neural network composed of pseudo-siamese network and generative adversarial net, and then train it with only separable states to construct non-linear witnesses for entanglement. It is shown via numerical examples, ranging from two-qubit to ten-qubit systems, that our network is able to achieve high detection accuracy which is above 97.5% on average.Moreover, it is capable of revealing rich structures of entanglement, such as partial entanglement among subsystems. Our results are readily applicable to the detection of other quantum resources such as Bell nonlocality and steerability, and thus our work could provide a powerful tool to extract quantum features hidden in multipartite quantum data.

Yiwei Chen, Yu Pan, Daoyi Dong

Quantum Language Models (QLMs) in which words are modelled as quantum superposition of sememes have demonstrated a high level of model transparency and good post-hoc interpretability. Nevertheless, in the current literature word sequences are basically modelled as a classical mixture of word states, which cannot fully exploit the potential of a quantum probabilistic description. A full quantum model is yet to be developed to explicitly capture the non-classical correlations within the word sequences. We propose a neural network model with a novel Entanglement Embedding (EE) module, whose function is to transform the word sequences into entangled pure states of many-body quantum systems. Strong quantum entanglement, which is the central concept of quantum information and an indication of parallelized correlations among the words, is observed within the word sequences. Numerical experiments show that the proposed QLM with EE (QLM-EE) achieves superior performance compared with the classical deep neural network models and other QLMs on Question Answering (QA) datasets. In addition, the post-hoc interpretability of the model can be improved by quantizing the degree of entanglement among the words.

Wanyue Li, Wen Kong, Yiwei Chen et al.

Fluorescein angiography can provide a map of retinal vascular structure and function, which is commonly used in ophthalmology diagnosis, however, this imaging modality may pose risks of harm to the patients. To help physicians reduce the potential risks of diagnosis, an image translation method is adopted. In this work, we proposed a conditional generative adversarial network(GAN) - based method to directly learn the mapping relationship between structure fundus images and fundus fluorescence angiography images. Moreover, local saliency maps, which define each pixel's importance, are used to define a novel saliency loss in the GAN cost function. This facilitates more accurate learning of small-vessel and fluorescein leakage features.