Multimodal Large Language Models (MLLMs) have recently sparked significant interest, which demonstrates emergent capabilities to serve as a general-purpose model for various vision-language tasks. However, existing methods mainly focus on limited types of instructions with a single image as visual context, which hinders the widespread availability of MLLMs. In this paper, we introduce the I4 benchmark to comprehensively evaluate the instruction following ability on complicated interleaved vision-language instructions, which involve intricate image-text sequential context, covering a diverse range of scenarios (e.g., visually-rich webpages/textbooks, lecture slides, embodied dialogue). Systematic evaluation on our I4 benchmark reveals a common defect of existing methods: the Visual Prompt Generator (VPG) trained on image-captioning alignment objective tends to attend to common foreground information for captioning but struggles to extract specific information required by particular tasks. To address this issue, we propose a generic and lightweight controllable knowledge re-injection module, which utilizes the sophisticated reasoning ability of LLMs to control the VPG to conditionally extract instruction-specific visual information and re-inject it into the LLM. Further, we introduce an annotation-free cross-attention guided counterfactual image training strategy to methodically learn the proposed module by collaborating a cascade of foundation models. Enhanced by the proposed module and training strategy, we present Cheetor, a Transformer-based MLLM that can effectively handle a wide variety of interleaved vision-language instructions and achieves state-of-the-art zero-shot performance across all tasks of I4, without high-quality multimodal instruction tuning data. Cheetor also exhibits competitive performance compared with state-of-the-art instruction tuned models on MME benchmark.
There is a rapidly-growing research interest in engaging users with multi-modal data for accurate user modeling on recommender systems. Existing multimedia recommenders have achieved substantial improvements by incorporating various modalities and devising delicate modules. However, when users decide to interact with items, most of them do not fully read the content of all modalities. We refer to modalities that directly cause users' behaviors as point-of-interests, which are important aspects to capture users' interests. In contrast, modalities that do not cause users' behaviors are potential noises and might mislead the learning of a recommendation model. Not surprisingly, little research in the literature has been devoted to denoising such potential noises due to the inaccessibility of users' explicit feedback on their point-of-interests. To bridge the gap, we propose a weakly-supervised framework based on contrastive learning for denoising multi-modal recommenders (dubbed Demure). In a weakly-supervised manner, Demure circumvents the requirement of users' explicit feedback and identifies the noises by analyzing the modalities of all interacted items from a given user.
Conventional multi-label classification (MLC) methods assume that all samples are fully labeled and identically distributed. Unfortunately, this assumption is unrealistic in large-scale MLC data that has long-tailed (LT) distribution and partial labels (PL). To address the problem, we introduce a novel task, Partial labeling and Long-Tailed Multi-Label Classification (PLT-MLC), to jointly consider the above two imperfect learning environments. Not surprisingly, we find that most LT-MLC and PL-MLC approaches fail to solve the PLT-MLC, resulting in significant performance degradation on the two proposed PLT-MLC benchmarks. Therefore, we propose an end-to-end learning framework: \textbf{CO}rrection $\rightarrow$ \textbf{M}odificat\textbf{I}on $\rightarrow$ balan\textbf{C}e, abbreviated as \textbf{\method{}}. Our bootstrapping philosophy is to simultaneously correct the missing labels (Correction) with convinced prediction confidence over a class-aware threshold and to learn from these recall labels during training. We next propose a novel multi-focal modifier loss that simultaneously addresses head-tail imbalance and positive-negative imbalance to adaptively modify the attention to different samples (Modification) under the LT class distribution. In addition, we develop a balanced training strategy by distilling the model's learning effect from head and tail samples, and thus design a balanced classifier (Balance) conditioned on the head and tail learning effect to maintain stable performance for all samples. Our experimental study shows that the proposed \method{} significantly outperforms general MLC, LT-MLC and PL-MLC methods in terms of effectiveness and robustness on our newly created PLT-MLC datasets.
Electronic Health Records (EHR) are generated from clinical routine care recording valuable information of broad patient populations, which provide plentiful opportunities for improving patient management and intervention strategies in clinical practice. To exploit the enormous potential of EHR data, a popular EHR data analysis paradigm in machine learning is EHR representation learning, which first leverages the individual patient's EHR data to learn informative representations by a backbone, and supports diverse health-care downstream tasks grounded on the representations. Unfortunately, such a paradigm fails to access the in-depth analysis of patients' relevance, which is generally known as cohort studies in clinical practice. Specifically, patients in the same cohort tend to share similar characteristics, implying their resemblance in medical conditions such as symptoms or diseases. In this paper, we propose a universal COhort Representation lEarning (CORE) framework to augment EHR utilization by leveraging the fine-grained cohort information among patients. In particular, CORE first develops an explicit patient modeling task based on the prior knowledge of patients' diagnosis codes, which measures the latent relevance among patients to adaptively divide the cohorts for each patient. Based on the constructed cohorts, CORE recodes the pre-extracted EHR data representation from intra- and inter-cohort perspectives, yielding augmented EHR data representation learning. CORE is readily applicable to diverse backbone models, serving as a universal plug-in framework to infuse cohort information into healthcare methods for boosted performance. We conduct an extensive experimental evaluation on two real-world datasets, and the experimental results demonstrate the effectiveness and generalizability of CORE.
Semi-supervised domain adaptation (SSDA) adapts a learner to a new domain by effectively utilizing source domain data and a few labeled target samples. It is a practical yet under-investigated research topic. In this paper, we analyze the SSDA problem from two perspectives that have previously been overlooked, and correspondingly decompose it into two \emph{key subproblems}: \emph{robust domain adaptation (DA) learning} and \emph{maximal cross-domain data utilization}. \textbf{(i)} From a causal theoretical view, a robust DA model should distinguish the invariant ``concept'' (key clue to image label) from the nuisance of confounding factors across domains. To achieve this goal, we propose to generate \emph{concept-invariant samples} to enable the model to classify the samples through causal intervention, yielding improved generalization guarantees; \textbf{(ii)} Based on the robust DA theory, we aim to exploit the maximal utilization of rich source domain data and a few labeled target samples to boost SSDA further. Consequently, we propose a collaboratively debiasing learning framework that utilizes two complementary semi-supervised learning (SSL) classifiers to mutually exchange their unbiased knowledge, which helps unleash the potential of source and target domain training data, thereby producing more convincing pseudo-labels. Such obtained labels facilitate cross-domain feature alignment and duly improve the invariant concept learning. In our experimental study, we show that the proposed model significantly outperforms SOTA methods in terms of effectiveness and generalisability on SSDA datasets.
Prompt tuning, a recently emerging paradigm, enables the powerful vision-language pre-training models to adapt to downstream tasks in a parameter -- and data -- efficient way, by learning the ``soft prompts'' to condition frozen pre-training models. Though effective, it is particularly problematic in the few-shot scenario, where prompt tuning performance is sensitive to the initialization and requires a time-consuming process to find a good initialization, thus restricting the fast adaptation ability of the pre-training models. In addition, prompt tuning could undermine the generalizability of the pre-training models, because the learnable prompt tokens are easy to overfit to the limited training samples. To address these issues, we introduce a novel Gradient-RegulAted Meta-prompt learning (GRAM) framework that jointly meta-learns an efficient soft prompt initialization for better adaptation and a lightweight gradient regulating function for strong cross-domain generalizability in a meta-learning paradigm using only the unlabeled image-text pre-training data. Rather than designing a specific prompt tuning method, our GRAM can be easily incorporated into various prompt tuning methods in a model-agnostic way, and comprehensive experiments show that GRAM brings about consistent improvement for them in several settings (i.e., few-shot learning, cross-domain generalization, cross-dataset generalization, etc.) over 11 datasets. Further, experiments show that GRAM enables the orthogonal methods of textual and visual prompt tuning to work in a mutually-enhanced way, offering better generalizability beyond the uni-modal prompt tuning methods.
Recommendation systems have shown great potential to solve the information explosion problem and enhance user experience in various online applications, which recently present two emerging trends: (i) Collaboration: single-sided model trained on-cloud (separate learning) to the device-cloud collaborative recommendation (collaborative learning). (ii) Real-time Dynamic: the network parameters are the same across all the instances (static model) to adaptive network parameters generation conditioned on the real-time instances (dynamic model). The aforementioned two trends enable the device-cloud collaborative and dynamic recommendation, which deeply exploits the recommendation pattern among cloud-device data and efficiently characterizes different instances with different underlying distributions based on the cost of frequent device-cloud communication. Despite promising, we argue that most of the communications are unnecessary to request the new parameters of the recommendation system on the cloud since the on-device data distribution are not always changing. To alleviate this issue, we designed a Intelligent DEvice-Cloud PArameter Request ModeL (IDEAL) that can be deployed on the device to calculate the request revenue with low resource consumption, so as to ensure the adaptive device-cloud communication with high revenue. We envision a new device intelligence learning task to implement IDEAL by detecting the data out-of-domain. Moreover, we map the user's real-time behavior to a normal distribution, the uncertainty is calculated by the multi-sampling outputs to measure the generalization ability of the device model to the current user behavior. Our experimental study demonstrates IDEAL's effectiveness and generalizability on four public benchmarks, which yield a higher efficient device-cloud collaborative and dynamic recommendation paradigm.
Temporal grounding is the task of locating a specific segment from an untrimmed video according to a query sentence. This task has achieved significant momentum in the computer vision community as it enables activity grounding beyond pre-defined activity classes by utilizing the semantic diversity of natural language descriptions. The semantic diversity is rooted in the principle of compositionality in linguistics, where novel semantics can be systematically described by combining known words in novel ways (compositional generalization). However, existing temporal grounding datasets are not carefully designed to evaluate the compositional generalizability. To systematically benchmark the compositional generalizability of temporal grounding models, we introduce a new Compositional Temporal Grounding task and construct two new dataset splits, i.e., Charades-CG and ActivityNet-CG. When evaluating the state-of-the-art methods on our new dataset splits, we empirically find that they fail to generalize to queries with novel combinations of seen words. We argue that the inherent structured semantics inside the videos and language is the crucial factor to achieve compositional generalization. Based on this insight, we propose a variational cross-graph reasoning framework that explicitly decomposes video and language into hierarchical semantic graphs, respectively, and learns fine-grained semantic correspondence between the two graphs. Furthermore, we introduce a novel adaptive structured semantics learning approach to derive the structure-informed and domain-generalizable graph representations, which facilitate the fine-grained semantic correspondence reasoning between the two graphs. Extensive experiments validate the superior compositional generalizability of our approach.
Understanding human emotions is a crucial ability for intelligent robots to provide better human-robot interactions. The existing works are limited to trimmed video-level emotion classification, failing to locate the temporal window corresponding to the emotion. In this paper, we introduce a new task, named Temporal Emotion Localization in videos~(TEL), which aims to detect human emotions and localize their corresponding temporal boundaries in untrimmed videos with aligned subtitles. TEL presents three unique challenges compared to temporal action localization: 1) The emotions have extremely varied temporal dynamics; 2) The emotion cues are embedded in both appearances and complex plots; 3) The fine-grained temporal annotations are complicated and labor-intensive. To address the first two challenges, we propose a novel dilated context integrated network with a coarse-fine two-stream architecture. The coarse stream captures varied temporal dynamics by modeling multi-granularity temporal contexts. The fine stream achieves complex plots understanding by reasoning the dependency between the multi-granularity temporal contexts from the coarse stream and adaptively integrates them into fine-grained video segment features. To address the third challenge, we introduce a cross-modal consensus learning paradigm, which leverages the inherent semantic consensus between the aligned video and subtitle to achieve weakly-supervised learning. We contribute a new testing set with 3,000 manually-annotated temporal boundaries so that future research on the TEL problem can be quantitatively evaluated. Extensive experiments show the effectiveness of our approach on temporal emotion localization. The repository of this work is at https://github.com/YYJMJC/Temporal-Emotion-Localization-in-Videos.
Content-Based Image Retrieval (CIR) aims to search for a target image by concurrently comprehending the composition of an example image and a complementary text, which potentially impacts a wide variety of real-world applications, such as internet search and fashion retrieval. In this scenario, the input image serves as an intuitive context and background for the search, while the corresponding language expressly requests new traits on how specific characteristics of the query image should be modified in order to get the intended target image. This task is challenging since it necessitates learning and understanding the composite image-text representation by incorporating cross-granular semantic updates. In this paper, we tackle this task by a novel \underline{\textbf{B}}ottom-up cr\underline{\textbf{O}}ss-modal \underline{\textbf{S}}emantic compo\underline{\textbf{S}}ition (\textbf{BOSS}) with Hybrid Counterfactual Training framework, which sheds new light on the CIR task by studying it from two previously overlooked perspectives: \emph{implicitly bottom-up composition of visiolinguistic representation} and \emph{explicitly fine-grained correspondence of query-target construction}. On the one hand, we leverage the implicit interaction and composition of cross-modal embeddings from the bottom local characteristics to the top global semantics, preserving and transforming the visual representation conditioned on language semantics in several continuous steps for effective target image search. On the other hand, we devise a hybrid counterfactual training strategy that can reduce the model's ambiguity for similar queries.