*Visual reprogramming* (VR) reuses pre-trained vision models for downstream image classification tasks by adding trainable noise patterns to inputs. When applied to vision-language models (e.g., CLIP), existing VR approaches follow the same pipeline used in vision models (e.g., ResNet, ViT), where ground-truth class labels are inserted into fixed text templates to guide the optimization of VR patterns. This label-based approach, however, overlooks the rich information and diverse attribute-guided textual representations that CLIP can exploit, which may lead to the misclassification of samples. In this paper, we propose ***Attr**ibute-based **V**isual **R**eprogramming* (AttrVR) for CLIP, utilizing ***des**criptive **attr**ibutes* (DesAttrs) and ***dist**inctive **attr**ibutes* (DistAttrs), which respectively represent common and unique feature descriptions for different classes. Besides, as images of the same class may reflect different attributes after VR, AttrVR iteratively refines patterns using the $k$-nearest DesAttrs and DistAttrs for each image sample, enabling more dynamic and sample-specific optimization. Theoretically, AttrVR is shown to reduce intra-class variance and increase inter-class separation. Empirically, it achieves superior performance in 12 downstream tasks for both ViT-based and ResNet-based CLIP. The success of AttrVR facilitates more effective integration of VR from unimodal vision models into vision-language models. Our code is available at https://github.com/tmlr-group/AttrVR.

Model reprogramming adapts pretrained models to downstream tasks by modifying only the input and output spaces. *Visual reprogramming* (VR) is one instance for vision tasks that adds a trainable noise pattern (i.e., a visual prompt) to input images to facilitate downstream classification. The existing VR approaches for CLIP train a single visual prompt using all descriptions of different downstream classes. However, the limited learning capacity may result in (1) a failure to capture diverse aspects of the descriptions (e.g., shape, color, and texture), and (2) a possible bias toward less informative attributes that do not help distinguish between classes. In this paper, we introduce a decoupling-and-reweighting framework. Our *decoupled visual prompts* (DVP) are optimized using descriptions grouped by explicit **c**au**se**s (DVP-cse) or unsupervised **cl**u**s**ters (DVP-cls). Then, we integrate the outputs of these visual prompts with a *probabilistic reweighting matrix* (PRM) that measures their contributions to each downstream class. Theoretically, DVP lowers the empirical risk bound. Experimentally, DVP outperforms baselines on average across 11 downstream datasets. Notably, the DVP-PRM integration enables insights into how individual visual prompts influence classification decisions, providing a probabilistic framework for understanding reprogramming.

*Diffusion-based purification* (DBP) methods aim to remove adversarial noise from the input sample by first injecting Gaussian noise through a forward diffusion process, and then recovering the clean example through a reverse generative process. In the above process, how much Gaussian noise is injected to the input sample is key to the success of DBP methods, which is controlled by a constant noise level $t*$ for all samples in existing methods. In this paper, we discover that an optimal $t*$ for each sample indeed could be different. Intuitively, the cleaner a sample is, the less the noise it should be injected, and vice versa. Motivated by this finding, we propose a new framework, called Sample-specific Score-aware Noise Injection (SSNI). Specifically, SSNI uses a pre-trained score network to estimate how much a data point deviates from the clean data distribution (i.e., score norms). Then, based on the magnitude of score norms, SSNI applies a reweighting function to adaptively adjust $t*$ for each sample, achieving sample-specific noise injections. Empirically, incorporating our framework with existing DBP methods results in a notable improvement in both accuracy and robustness on CIFAR-10 and ImageNet-1K, highlighting the necessity to allocate *distinct noise levels to different samples* in DBP methods. Our code is available at: https://github.com/tmlr-group/SSNI.

Visual reprogramming (VR) is a prompting technique that aims to re-purpose a pre-trained model (e.g., a classifier on ImageNet) to target tasks (e.g., medical data prediction) by learning a small-scale pattern added into input images instead of tuning considerable parameters within the model. The location of the pattern within input samples is usually determined by a pre-defined mask shared across all samples. In this paper, we show that the shared mask potentially limits VR’s generalization and increases its approximation error due to the lack of sample-level adaptation. Motivated by this finding, we design a new framework for VR called sample-specific multi-channel masks (SMM). Specifically, SMM employs a lightweight ConvNet and patch-wise interpolation to generate sample-specific three-channel masks instead of a shared and pre-defined mask. Since we generate different masks for individual samples, SMM is theoretically shown to reduce approximation error for the target tasks compared with existing state-of-the-art VR methods. We also empirically demonstrate its performance gain on both ResNet and ViT. The success of SMM further highlights the broader applicability of VR in leveraging the latent knowledge of pre-trained models for various target tasks. Our code is available at https://github.com/tmlr-group/SMM.

*Visual reprogramming* (VR) leverages the intrinsic capabilities of pretrained vision models by adapting their input or output interfaces to solve downstream tasks whose labels (i.e., downstream labels) might be totally different from the labels associated with the pretrained models (i.e., pretrained labels). When adapting the output interface, label mapping methods transform the pretrained labels to downstream labels by establishing a gradient-free one-to-one correspondence between the two sets of labels. However, in this paper, we reveal that one-to-one mappings may overlook the complex relationship between pretrained and downstream labels. Motivated by this observation, we propose a ***B**ayesian-guided **L**abel **M**apping* (BLM) method. BLM constructs an iteratively-updated probabilistic label mapping matrix, with each element quantifying a pairwise relationship between pretrained and downstream labels. The assignment of values to the constructed matrix is guided by Bayesian conditional probability, considering the joint distribution of the downstream labels and the labels predicted by the pretrained model on downstream samples. Experiments conducted on both pretrained vision models (e.g., ResNeXt) and vision-language models (e.g., CLIP) demonstrate the superior performance of BLM over existing label mapping methods. The success of BLM also offers a probabilistic lens through which to understand and analyze the effectiveness of VR. Our code is available at https://github.com/tmlr-group/BayesianLM.

Cross-Domain Recommendation~(CDR) seeks to enable effective knowledge transfer across domains. Most existing works rely on either representation alignment or transformation bridges, but they come with shortcomings regarding identifiability of domain-shared and domain-specific latent factors. Specifically, while CDR describes user representations as a joint distribution over two domains, these methods fail to account for its joint identifiability as they primarily fixate on the marginal distribution within a particular domain. Such a failure may overlook the conditionality between two domains and how it contributes to latent factor disentanglement, leading to negative transfer when domains are weakly correlated. In this study, we explore what should and should not be transferred in cross-domain user representations from a causality perspective. We propose a Hierarchical causal subspace disentanglement approach to explore the Joint IDentifiability of cross-domain joint distribution, termed HJID, to preserve domain-specific behaviors from domain-shared factors. HJID abides by the feature hierarchy and divides user representations into generic shallow subspace and domain-oriented deep subspaces. We first encode the generic pattern in the shallow subspace by minimizing the Maximum Mean Discrepancy of initial layer activation. Then, to dissect how domain-oriented latent factors are encoded in deeper layers activation, we construct a cross-domain causality-based data generation graph, which identifies cross-domain consistent and domain-specific components, adhering to the Minimal Change principle. This allows HJID to maintain stability whilst discovering unique factors for different domains, all within a generative framework of invertible transformations that guarantee the joint identifiability. With experiments on real-world datasets, we show that HJID outperforms SOTA methods on both strong- and weak-correlation CDR tasks.

Tremendous efforts have been put forth on predicting pedestrian trajectory with generative models to accommodate uncertainty and multi-modality in human behaviors. An individual’s inherent uncertainty, e.g., change of destination, can be masked by complex patterns resulting from the movements of interacting pedestrians. However, latent variable-based generative models often entangle such uncertainty with complexity, leading to limited either latent expressivity or predictive diversity. In this work, we propose to separately model these two factors by implicitly deriving a flexible latent representation to capture intricate pedestrian movements, while integrating predictive uncertainty of individuals with explicit bivariate Gaussian mixture densities over their future locations. More specifically, we present a model-agnostic uncertainty-aware pedestrian trajectory prediction framework, parameterizing sufficient statistics for the mixture of Gaussians that jointly comprise the multi-modal trajectories. We further estimate these parameters of interest by approximating a denoising process that progressively recovers pedestrian movements from noise. Unlike previous studies, we translate the predictive stochasticity to explicit distributions, allowing it to readily generate plausible future trajectories indicating individuals’ self-uncertainty. Moreover, our framework is compatible with different neural net architectures. We empirically show the performance gains over state-of-the-art even with lighter backbones, across most scenes on two public benchmarks.

We study the problem of restoring visual stimuli from electroencephalography (EEG) signals. Using a supervised classification-then-generation framework, the reconstruction-based approaches learn the mapping between distributions of two modalities but fail to reproduce the exact visual stimulus. Instead, we propose a self-supervised cross-modal retrieval paradigm that seeks instance-level alignment by maximizing the mutual information between the EEG encoding and associated visual stimulus. We demonstrate the threefold advantages of the self-supervised retrieval over supervised reconstruction on the largest visual-evoked EEG dataset with two evaluation protocols. First, it restores the exact visual stimulus without accessing the image class information, which was not possible with previous approaches. Second, it produces more recognizable results than generated ones and bypasses the challenge of training an image generator. Finally, it illustrates the benefits of self-supervision over supervised models in handling open-set data.

Neural Processes (NPs) are a family of supervised density estimators devoted to probabilistic function approximation with meta-learning. Despite extensive research on the subject, the absence of a unified framework for NPs leads to varied architectural solutions across diverse studies. This non-consensus poses challenges to reproducing and benchmarking different NPs. Moreover, existing codebases mainly prioritize generative density estimation, yet rarely consider expanding the capability of NPs to self-supervised representation learning, which however has gained growing importance in data mining applications. To this end, we present NP-SSL, a modular and configurable framework with built-in support, requiring minimal effort to 1) implement classical NPs architectures; 2) customize specific components; 3) integrate hybrid training scheme (e.g., contrastive); and 4) extend NPs to act as a self-supervised learning toolkit, producing latent representations of data, and facilitating diverse downstream predictive tasks. To illustrate, we discuss a case study that applies NP-SSL to model time-series data. We interpret that NP-SSL can handle different predictive tasks such as imputation and forecasting, by a simple switch in data samplings, without significant change to the underlying structure. We hope this study can reduce the workload of future research on leveraging NPs to tackle more a broader range of real-world data mining applications. Code and documentation are at https://github.com/zyecs/NP-SSL.

Learning accurate cross-domain preference mappings in the absence of overlapped users/items has presented a persistent challenge in Non-overlapping Cross-domain Recommendation (NOCDR). Despite the efforts made in previous studies to address NOCDR, several limitations still exist. Specifically, 1) while some approaches substitute overlapping users/items with overlapping behaviors, they cannot handle NOCDR scenarios where such auxiliary information is unavailable; 2) often, cross-domain preference mapping is modeled by learning deterministic explicit representation matchings between sampled users in two domains. However, this can be biased due to individual preferences and thus fails to incorporate preference continuity and universality of the general population. In light of this, we assume that despite the scattered nature of user behaviors, there exists a consistent latent preference distribution shared among common people. Modeling such distributions further allows us to capture the continuity in user behaviors within each domain and discover preference invariance across domains. To this end, we propose a Distributional domain-invariant Preference Matching method for non-overlapping Cross-Domain Recommendation (DPMCDR). For each domain, we hierarchically approximate a posterior of domain-level preference distribution with empirical evidence derived from user-item interactions. Next, we aim to build distributional implicit matchings between the domain-level preferences of two domains. This process involves mapping them to a shared latent space and seeking a consensus on domain-invariant preference by minimizing the distance between their distributional representations therein. In this way, we can identify the alignment of two non-overlapping domains if they exhibit similar patterns of domain-invariant preference. Experiments on real-world datasets demonstrate that DPMCDR outperforms the state-of-the-art approaches with a range of evaluation metrics.

Recent sequential recommendation models rely increasingly on consecutive short-term user-item interaction sequences to model user interests. These approaches have, however, raised concerns about both short- and long-term interests. (1) short-term: interaction sequences may not result from a monolithic interest, but rather from several intertwined interests, even within a short period of time, resulting in their failures to model skip behaviors; (2) long-term: interaction sequences are primarily observed sparsely at discrete intervals, other than consecutively over the long run. This renders difficulty in inferring long-term interests, since only discrete interest representations can be derived, without taking into account interest dynamics across sequences. In this study, we address these concerns by learning (1) multi-scale representations of short-term interests; and (2) dynamics-aware representations of long-term interests. To this end, we present an Interest Dynamics modeling framework using generative Neural Processes, coined IDNP, to model user interests from a functional perspective. IDNP learns a global interest function family to define each user's long-term interest as a function instantiation, manifesting interest dynamics through function continuity. Specifically, IDNP first encodes each user's short-term interactions into multi-scale representations, which are then summarized as user context. By combining latent global interest with user context, IDNP then reconstructs long-term user interest functions and predicts interactions at upcoming query timestep. Moreover, IDNP can model such interest functions even when interaction sequences are limited and non-consecutive. Extensive experiments on four real-world datasets demonstrate that our model outperforms the state-of-the-art on various evaluation metrics.

Conditional Neural Processes (CNPs) bridge neural net-works with probabilistic inference to approximate functions of Stochastic Processes under meta-learning settings. Given a batch of non-i.i.d function instantiations, CNPs are jointly optimized for in-instantiation observation prediction and cross-instantiation meta-representation adaptation within a generative reconstruction pipeline. There can be a challenge in tying together such two targets when the distri-bution of function observations scales to high-dimensional and noisy spaces. Instead, noise contrastive estimation might be able to provide more robust representations by learning distributional matching objectives to combat such inherent limitation of generative models. In light of this, we propose to equip CNPs by 1) aligning prediction with en-coded ground-truth observation, and 2) decoupling meta-representation adaptation from generative reconstruction. Specifically, two auxiliary contrastive branches are set up hierarchically, namely in-instantiation temporal contrastive learning (TCL) and cross-instantiation function contrastive learning (FCL), to facilitate local predictive alignment and global function consistency, respectively. We empirically show that TCL captures high-level abstraction of obser-vations, whereas FCL helps identify underlying functions, which in turn provides more efficient representations. Our model outperforms other CNPs variants when evaluating function distribution reconstruction and parameter identifi-cation across 1D, 2D and high-dimensional time-series.

Social recommendation with Graph Neural Networks(GNNs) learns to represent cold users by fusing user-user social relations with user-item interactions, thereby alleviating the cold-start problem associated with recommender systems. Despite being well adapted to social relations and user-item interactions, these supervised models are still susceptible to popularity bias. Contrastive learning helps resolve this dilemma by identifying the properties that distinguish positive from negative samples. In its previous combinations with recommender systems, social relationships and cold-start cases in this context are not considered. Also, they primarily focus on collaborative features between users and items, leaving the similarity between items under-utilized. In this work, we propose socially-aware dual contrastive learning for cold-start recommendation, where cold users can be modeled in the same way as warm users. To take full advantage of social relations, we create dynamic node embeddings for each user by aggregating information from different neighbors according to each different query item, in the form of user-item pairs. We further design a dual-branch self-supervised contrastive objective to account for user-item collaborative features and item-item mutual information, respectively. On one hand, our framework eliminates popularity bias with proper negative sampling in contrastive learning, without extra ground-truth supervision. On the other hand, we extend previous contrastive learning methods to provide a solution to cold-start problem with social relations included. Extensive experiments on two real-world social recommendation datasets demonstrate its effectiveness.