Large Multimodal Models (LMMs) face notable challenges when encountering multimodal knowledge conflicts, particularly under retrieval-augmented generation (RAG) frameworks, where the contextual information from external sources may contradict the model’s internal parametric knowledge, leading to unreliable outputs. However, existing benchmarks fail to reflect such realistic conflict scenarios. Most focus solely on intra-memory conflicts, while context-memory and inter-context conflicts remain largely unaddressed. Furthermore, commonly used factual knowledge-based evaluations are often overlooked, and existing datasets lack a thorough investigation into conflict detection capabilities.To bridge this gap, we propose MMKC-Bench, a benchmark designed to evaluate factual knowledge conflicts in both context-memory and inter-context scenarios. MMKC-Bench encompasses four types of multimodal knowledge conflicts and includes 1,881 knowledge instances and 3,997 images across 32 broad types, collected through automated pipelines with human verification. We evaluate four representative series of LMMs on both model behavior analysis and conflict detection tasks. Our findings show that while current LMMs are capable of recognizing knowledge conflicts, they tend to favor internal parametric knowledge over external evidence. We hope MMKC-Bench will foster further research in multimodal knowledge conflict and enhance the development of multimodal RAG systems.

Most explanation methods are designed in an empirical manner, so exploring whether there exists a first-principles explanation of a deep neural network (DNN) becomes the next core scientific problem in explainable artificial intelligence (XAI). Although it is still an open problem, in this paper, we discuss whether the interaction-based explanation can serve as the first-principles explanation of a DNN. The strong explanatory power of interaction theory comes from the following aspects: (1) it establishes a new axiomatic system to quantify the decision-making logic of a DNN into a set of symbolic interaction concepts; (2) it simultaneously explains various deep learning phenomena, such as generalization power, adversarial sensitivity, representation bottleneck, and learning dynamics; (3) it provides mathematical tools that uniformly explain the mechanisms of various empirical attribution methods and empirical adversarial-transferability-boosting methods; (4) it explains the extremely complex learning dynamics of a DNN by analyzing the two-phase dynamics of interaction complexity, which further reveals the internal mechanism of why and how the generalization power/adversarial sensitivity of a DNN changes during the learning process.

This paper focuses on the problem of preventing information leakage in neural networks, i.e., assuming that attackers have obtained intermediate-layer features of a neural network, and preventing attackers from inverting these features to the input with private information. We propose a generic method to slightly revise each arbitrary traditional neural network into a multiary-valued rotation-equivariant neural network (RENN) for preventing information leakage. Specifically, we convert real-valued features in the network into multi-ary features, and each element in the feature vector is a multi-ary number. We hide the input information into a certain phase of the multi-ary feature, and rotate the multi-ary feature for attribute obfuscation in the encryption process. The rotation axis and angle can be considered as the private key. In this way, even when attackers have obtained network parameters and intermediate-layer features, they still cannot extract input information without knowing the rotation information. More crucially, the encryption operation does not damage the spatial correlations between features, so that the encrypted features can be easily processed by convolution operations in the neural network without difficulties. In order to implement successful encryption and decryption, the RENN is designed to satisfy the rotation equivariance property. To this end, we propose a set of rules to revise classic operations in the neural network to ensure the rotation equivariance property. Besides, we prove that the $d$d-ary RENN is downward compatible with the $d^{\prime }$d'-ary RENN when $d^{\prime }< d$d'

This study aims to prove the emergence of symbolic concepts (or more precisely, sparse primitive inference patterns) in well-trained deep neural networks (DNNs). Specifically, we prove the following three conditions for the emergence. (i) The high-order derivatives of the network output with respect to the input variables are all zero. (ii) The DNN can be used on occluded samples, and when the input sample is less occluded, the DNN will yield higher confidence. (iii) The confidence of the DNN does not significantly degrade on occluded samples. These conditions are quite common, and we prove that under these conditions, the DNN will only encode a relatively small number of sparse interactions between input variables. Moreover, we can consider such interactions as symbolic primitive inference patterns encoded by a DNN, because we show that inference scores of the DNN on an exponentially large number of randomly masked samples can always be well mimicked by numerical effects of just a few interactions.

This study proves the two-phase dynamics of a deep neural network (DNN) learning interactions. Despite the long disappointing view of the faithfulness of post-hoc explanation of a DNN, a series of theorems have been proven [27] in recent years to show that for a given input sample, a small set of interactions between input variables can be considered as primitive inference patterns that faithfully represent a DNN's detailed inference logic on that sample. Particularly, Zhang et al. [41] have observed that various DNNs all learn interactions of different complexities in two distinct phases, and this two-phase dynamics well explains how a DNN changes from under-fitting to over-fitting. Therefore, in this study, we mathematically prove the two-phase dynamics of interactions, providing a theoretical mechanism for how the generalization power of a DNN changes during the training process. Experiments show that our theory well predicts the real dynamics of interactions on different DNNs trained for various tasks.

This study proposes a set of generic rules to revise existing neural networks for 3D point cloud processing to rotation-equivariant quaternion neural networks (REQNNs), in order to make feature representations of neural networks to be rotation-equivariant and permutation-invariant. Rotation equivariance of features means that the feature computed on a rotated input point cloud is the same as applying the same rotation transformation to the feature computed on the original input point cloud. We find that the rotation-equivariance of features is naturally satisfied, if a neural network uses quaternion features. Interestingly, we prove that such a network revision also makes gradients of features in the REQNN to be rotation-equivariant w.r.t. inputs, and the training of the REQNN to be rotation-invariant w.r.t. inputs. Besides, permutation-invariance examines whether the intermediate-layer features are invariant, when we reorder input points. We also evaluate the stability of knowledge representations of REQNNs, and the robustness of REQNNs to adversarial rotation attacks. Experiments have shown that REQNNs outperform traditional neural networks in both terms of classification accuracy and robustness on rotated testing samples.

In this paper, we focus on mean-field variational Bayesian Neural Networks (BNNs) and explore the representation capacity of such BNNs by investigating which types of concepts are less likely to be encoded by the BNN. It has been observed and studied that a relatively small set of interactive concepts usually emerge in the knowledge representation of a sufficiently-trained neural network, and such concepts can faithfully explain the network output. Based on this, our study proves that compared to standard deep neural networks (DNNs), it is less likely for BNNs to encode complex concepts. Experiments verify our theoretical proofs. Note that the tendency to encode less complex concepts does not necessarily imply weak representation power, considering that complex concepts exhibit low generalization power and high adversarial vulnerability. The code is available at https://github.com/sjtu-xai-lab/BNN-concepts.

This paper theoretically explains the intuition that simple concepts are more likely to be learned by deep neural networks (DNNs) than complex concepts. In fact, recent studies have observed [24, 15] and proved [26] the emergence of interactive concepts in a DNN, i.e., it is proven that a DNN usually only encodes a small number of interactive concepts, and can be considered to use their interaction effects to compute inference scores. Each interactive concept is encoded by the DNN to represent the collaboration between a set of input variables. Therefore, in this study, we aim to theoretically explain that interactive concepts involving more input variables (i.e., more complex concepts) are more difficult to learn. Our finding clarifies the exact conceptual complexity that boosts the learning difficulty.

In this paper, we evaluate the quality of knowledge representations encoded in deep neural networks (DNNs) for 3D point cloud processing. We propose a method to disentangle the overall model vulnerability into the sensitivity to the rotation, the translation, the scale, and local 3D structures. Besides, we also propose metrics to evaluate the spatial smoothness of encoding 3D structures, and the representation complexity of the DNN. Based on such analysis, experiments expose representation problems with classic DNNs, and explain the utility of the adversarial training. The code will be released when this paper is accepted.