The exorbitant cost of training Large Language Models (LLMs) makes it essential to protect the models from illegal copying and unauthorized usage. Recent attempts at LLM protection utilize black-box watermarking schemes, which embed distinctive input-output mapping (i.e., trigger set) directly into the models. However, most of them construct trigger inputs by injecting abnormal characters into normal text, which can easily be filtered out by unauthorized users, leading to a failure in watermark verification. In this paper, we propose a novel filtering-resistant LLM watermarking scheme, which takes advantage of imperceptible text styles to trigger the watermark. To achieve this, we adopt a trigger generation network to transform normal text into stylized sentences, which are assigned a specific watermarking label to build the trigger set. We then fine-tune the LLMs on both the trigger sets and clean samples for watermark embedding and performance stabilization. To boost watermark accuracy, we further propose a feature separation loss term to distinguish between normal and trigger inputs. Experimental results indicate the effectiveness of our proposed scheme for resisting the filtering attack.

Multimodal Dialogue agents are often required to respond to conversation history using both textual and visual content. Even though current dialogue studies predominantly strive to generate natural texts or images, they fall short in considering the relevance of multimodal responses within a dialogue context, consequently confining agents from making prudent choices based on multiple alternatives and their associated relevance scores for decision-making. In this paper, we present a bidirectional multimodal dialogue framework that skillfully combines the forward generation of multiple text and image response candidates with reverse selection guided by relevance scores evaluated on dialogue context, facilitating agents in selecting the most suitable multimodal responses. Specifically, the forward generation aspect of our framework leverages a stage-wise approach, first producing textual replies and composite visual descriptions from the dialogue context, followed by the generation of visual responses aligned with the descriptions. In the reverse selection process, visual responses are translated into tangible descriptive texts that, in conjunction with textual responses, are inversely tied back to the dialogue context for relevance assessment, assigning a reference score to each multimodal response candidate to assist the intelligent agent in making informed decisions. Experimental outcomes demonstrate that our proposed bidirectional dialogue response framework markedly elevates performance in both automatic and human evaluations, yielding a range of contextually fitting multimodal responses for selection.

Recent advances have emphasized the importance of meme stickers in open-domain dialogue systems.However, previous studies overlook the one-to-many issue that a single sticker could represent various emotions in different dialogue contexts.Additionally, they require retraining the model for new stickers which did not appear in previous training.To address the above issues, we propose in this paper an Emotion-Aware and Efficient Meme Sticker Dialogue generation framework.In the framework, we design an Emotion Adaptive Prompt to capture the emotional cues from the dialogue history, which is sent to an Emotion-Aware Fusion Decoder to guide the generation of text responses and to a meme sticker selector to choose the corresponding sticker.Furthermore, to improve the stickers' selection efficiency, we further incorporate the few-shot learning strategy into the proposed framework to avoid extensive model retraining for unseen meme stickers.Through extensive experiments, we demonstrate the superior performance of the proposed E 2 MSD compared to existing methods regarding the quality of response generation and the efficiency of meme sticker retrieval.

Text-to-image models have shown surprising performance in high-quality image generation, while also raising intensified concerns about the unauthorized usage of personal dataset in training and personalized fine-tuning. Recent approaches, embedding watermarks, introducing perturbations, and inserting backdoors into datasets, rely on adding minor information vulnerable to adversarial training, limiting their ability to detect unauthorized data usage. In this paper, we introduce a novel implicit Zero-Watermarking scheme that first utilizes the disentangled style domain to detect unauthorized dataset usage in text-to-image models. Specifically, our approach generates the watermark from the disentangled style domain, enabling self-generalization and mutual exclusivity within the style domain anchored by protected units. The domain achieves the maximum concealed offset of probability distribution through both the injection of identifier $z$ and dynamic contrastive learning, facilitating the structured delineation of dataset copyright boundaries for multiple sources of styles and contents. Additionally, we introduce the concept of watermark distribution to establish a verification mechanism for copyright ownership of hybrid or partial infringements, addressing deficiencies in the traditional mechanism of dataset copyright ownership for AI mimicry. Notably, our method achieves one-sample verification for copyright ownership in AI mimic generations. The code is available at: [https://github.com/Hlufies/ZWatermarking](https://github.com/Hlufies/ZWatermarking)

We present a simple but effective pixel-level self-supervised distillation framework friendly to dense prediction tasks. Our method, called Pixel-Wise Contrastive Distillation (PCD), distills knowledge by attracting the corresponding pixels from student’s and teacher’s output feature maps. PCD includes a novel design called SpatialAdaptor which "reshapes" a part of the teacher network while preserving the distribution of its output features. Our ablation experiments suggest that this reshaping behavior enables more informative pixel-to-pixel distillation. Moreover, we utilize a plug-in multi-head self-attention module that explicitly relates the pixels of student’s feature maps to enhance the effective receptive field, leading to a more competitive student. PCD outperforms previous self-supervised distillation methods on various dense prediction tasks. A backbone of ResNet-18-FPN distilled by PCD achieves 37.4 APbboxand 34.0 APmaskon COCO dataset using the detector of Mask R-CNN. We hope our study will inspire future research on how to pre-train a small model friendly to dense prediction tasks in a self-supervised fashion.