We demonstrate how images and sounds can be used for indirect prompt and instruction injection in multi-modal LLMs. An attacker generates an adversarial perturbation corresponding to the prompt and blends it into an image or audio recording. When the user asks the (unmodified, benign) model about the perturbed image or audio, the perturbation steers the model to output the attacker-chosen text and/or make the subsequent dialog follow the attacker's instruction. We illustrate this attack with several proof-of-concept examples targeting LLaVa and PandaGPT.
相關內容
Stochastic Computing (SC) is an unconventional computing paradigm processing data in the form of random bit-streams. The accuracy and energy efficiency of SC systems highly depend on the stochastic number generator (SNG) unit that converts the data from conventional binary to stochastic bit-streams. Recent work has shown significant improvement in the efficiency of SC systems by employing low-discrepancy (LD) sequences such as Sobol and Halton sequences in the SNG unit. Still, the usage of many well-known random sequences for SC remains unexplored. This work studies some new random sequences for potential application in SC. Our design space exploration proposes a promising random number generator for accurate and energy-efficient SC. We propose P2LSG, a low-cost and energy-efficient Low-discrepancy Sequence Generator derived from Powers-of-2 VDC (Van der Corput) sequences. We evaluate the performance of our novel bit-stream generator for two SC image and video processing case studies: image scaling and scene merging. For the scene merging task, we propose a novel SC design for the first time. Our experimental results show higher accuracy and lower hardware cost and energy consumption compared to the state-of-the-art.
We propose MAMo, a novel memory and attention frame-work for monocular video depth estimation. MAMo can augment and improve any single-image depth estimation networks into video depth estimation models, enabling them to take advantage of the temporal information to predict more accurate depth. In MAMo, we augment model with memory which aids the depth prediction as the model streams through the video. Specifically, the memory stores learned visual and displacement tokens of the previous time instances. This allows the depth network to cross-reference relevant features from the past when predicting depth on the current frame. We introduce a novel scheme to continuously update the memory, optimizing it to keep tokens that correspond with both the past and the present visual information. We adopt attention-based approach to process memory features where we first learn the spatio-temporal relation among the resultant visual and displacement memory tokens using self-attention module. Further, the output features of self-attention are aggregated with the current visual features through cross-attention. The cross-attended features are finally given to a decoder to predict depth on the current frame. Through extensive experiments on several benchmarks, including KITTI, NYU-Depth V2, and DDAD, we show that MAMo consistently improves monocular depth estimation networks and sets new state-of-the-art (SOTA) accuracy. Notably, our MAMo video depth estimation provides higher accuracy with lower latency, when omparing to SOTA cost-volume-based video depth models.
Diffusion models (DMs) have made significant progress in the fields of image, audio, and video generation. One downside of DMs is their slow iterative process. Recent algorithms for fast sampling are designed from the perspective of differential equations. However, in higher-order algorithms based on Taylor expansion, estimating the derivative of the score function becomes intractable due to the complexity of large-scale, well-trained neural networks. Driven by this motivation, in this work, we introduce the recursive difference (RD) method to calculate the derivative of the score function in the realm of DMs. Based on the RD method and the truncated Taylor expansion of score-integrand, we propose SciRE-Solver with the convergence order guarantee for accelerating sampling of DMs. To further investigate the effectiveness of the RD method, we also propose a variant named SciREI-Solver based on the RD method and exponential integrator. Our proposed sampling algorithms with RD method attain state-of-the-art (SOTA) FIDs in comparison to existing training-free sampling algorithms, across both discrete-time and continuous-time pre-trained DMs, under various number of score function evaluations (NFE). Remarkably, SciRE-Solver using a small NFEs demonstrates promising potential to surpass the FID achieved by some pre-trained models in their original papers using no fewer than $1000$ NFEs. For example, we reach SOTA value of $2.40$ FID with $100$ NFE for continuous-time DM and of $3.15$ FID with $84$ NFE for discrete-time DM on CIFAR-10, as well as of $2.17$ (2.02) FID with $18$ (50) NFE for discrete-time DM on CelebA 64$\times$64.
Foley sound synthesis refers to the creation of authentic, diegetic sound effects for media, such as film or radio. In this study, we construct a neural Foley synthesizer capable of generating mono-audio clips across seven predefined categories. Our approach introduces multiple enhancements to existing models in the text-to-audio domain, with the goal of enriching the diversity and acoustic characteristics of the generated foleys. Notably, we utilize a pre-trained encoder that retains acoustical and musical attributes in intermediate embeddings, implement class-conditioning to enhance differentiability among foley classes in their intermediate representations, and devise an innovative transformer-based architecture for optimizing self-attention computations on very large inputs without compromising valuable information. Subsequent to implementation, we present intermediate outcomes that surpass the baseline, discuss practical challenges encountered in achieving optimal results, and outline potential pathways for further research.
Markov games model interactions among multiple players in a stochastic, dynamic environment. Each player in a Markov game maximizes its expected total discounted reward, which depends upon the policies of the other players. We formulate a class of Markov games, termed affine Markov games, where an affine reward function couples the players' actions. We introduce a novel solution concept, the soft-Bellman equilibrium, where each player is boundedly rational and chooses a soft-Bellman policy rather than a purely rational policy as in the well-known Nash equilibrium concept. We provide conditions for the existence and uniqueness of the soft-Bellman equilibrium and propose a nonlinear least-squares algorithm to compute such an equilibrium in the forward problem. We then solve the inverse game problem of inferring the players' reward parameters from observed state-action trajectories via a projected-gradient algorithm. Experiments in a predator-prey OpenAI Gym environment show that the reward parameters inferred by the proposed algorithm outperform those inferred by a baseline algorithm: they reduce the Kullback-Leibler divergence between the equilibrium policies and observed policies by at least two orders of magnitude.
Various precoders have been recently studied by the wireless community to combat the channel fading effects. Two prominent precoders are implemented with the discrete Fourier transform (DFT) and Walsh-Hadamard transform (WHT). The WHT precoder is implemented with less complexity since it does not need complex multiplications. Also, spreading can be applied sparsely to decrease the transceiver complexity, leading to sparse DFT (SDFT) and sparse Walsh-Hadamard (SWH). Another relevant topic is the design of iterative receivers that deal with inter-symbol-interference (ISI). In particular, many detectors based on expectation propagation (EP) have been proposed recently for channels with high levels of ISI. An alternative is the maximum a-posterior (MAP) detector, although it leads to unfeasible high complexity in many cases. In this paper, we provide a relatively low-complexity \textcolor{black}{computation} of the MAP detector for the SWH. We also propose two \textcolor{black}{feasible methods} based on the Log-MAP and Max-Log-MAP. Additionally, the DFT, SDFT and SWH precoders are compared using an EP-based receiver with one-tap FD equalization. Lastly, SWH-Max-Log-MAP is compared to the (S)DFT with EP-based receiver in terms of performance and complexity. The results show that the proposed SWH-Max-Log-MAP has a better performance and complexity trade-off for QPSK and 16-QAM under highly selective channels, but has unfeasible complexity for higher QAM orders.
Recently Transformer and Convolution neural network (CNN) based models have shown promising results in EEG signal processing. Transformer models can capture the global dependencies in EEG signals through a self-attention mechanism, while CNN models can capture local features such as sawtooth waves. In this work, we propose an end-to-end neural epilepsy detection model, EENED, that combines CNN and Transformer. Specifically, by introducing the convolution module into the Transformer encoder, EENED can learn the time-dependent relationship of the patient's EEG signal features and notice local EEG abnormal mutations closely related to epilepsy, such as the appearance of spikes and the sprinkling of sharp and slow waves. Our proposed framework combines the ability of Transformer and CNN to capture different scale features of EEG signals and holds promise for improving the accuracy and reliability of epilepsy detection. Our source code will be released soon on GitHub.
Existing knowledge graph (KG) embedding models have primarily focused on static KGs. However, real-world KGs do not remain static, but rather evolve and grow in tandem with the development of KG applications. Consequently, new facts and previously unseen entities and relations continually emerge, necessitating an embedding model that can quickly learn and transfer new knowledge through growth. Motivated by this, we delve into an expanding field of KG embedding in this paper, i.e., lifelong KG embedding. We consider knowledge transfer and retention of the learning on growing snapshots of a KG without having to learn embeddings from scratch. The proposed model includes a masked KG autoencoder for embedding learning and update, with an embedding transfer strategy to inject the learned knowledge into the new entity and relation embeddings, and an embedding regularization method to avoid catastrophic forgetting. To investigate the impacts of different aspects of KG growth, we construct four datasets to evaluate the performance of lifelong KG embedding. Experimental results show that the proposed model outperforms the state-of-the-art inductive and lifelong embedding baselines.
Answering questions that require reading texts in an image is challenging for current models. One key difficulty of this task is that rare, polysemous, and ambiguous words frequently appear in images, e.g., names of places, products, and sports teams. To overcome this difficulty, only resorting to pre-trained word embedding models is far from enough. A desired model should utilize the rich information in multiple modalities of the image to help understand the meaning of scene texts, e.g., the prominent text on a bottle is most likely to be the brand. Following this idea, we propose a novel VQA approach, Multi-Modal Graph Neural Network (MM-GNN). It first represents an image as a graph consisting of three sub-graphs, depicting visual, semantic, and numeric modalities respectively. Then, we introduce three aggregators which guide the message passing from one graph to another to utilize the contexts in various modalities, so as to refine the features of nodes. The updated nodes have better features for the downstream question answering module. Experimental evaluations show that our MM-GNN represents the scene texts better and obviously facilitates the performances on two VQA tasks that require reading scene texts.
We propose a novel single shot object detection network named Detection with Enriched Semantics (DES). Our motivation is to enrich the semantics of object detection features within a typical deep detector, by a semantic segmentation branch and a global activation module. The segmentation branch is supervised by weak segmentation ground-truth, i.e., no extra annotation is required. In conjunction with that, we employ a global activation module which learns relationship between channels and object classes in a self-supervised manner. Comprehensive experimental results on both PASCAL VOC and MS COCO detection datasets demonstrate the effectiveness of the proposed method. In particular, with a VGG16 based DES, we achieve an mAP of 81.7 on VOC2007 test and an mAP of 32.8 on COCO test-dev with an inference speed of 31.5 milliseconds per image on a Titan Xp GPU. With a lower resolution version, we achieve an mAP of 79.7 on VOC2007 with an inference speed of 13.0 milliseconds per image.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191