Large Language Models (LLMs) have proven powerful, but the risk of privacy leakage remains a significant concern. Traditional privacy-preserving methods, such as Differential Privacy and Homomorphic Encryption, are inadequate for black-box API-only settings, demanding either model transparency or heavy computational resources. We propose Prompt2Forget (P2F), the first framework designed to tackle the LLM local privacy challenge by teaching LLM to forget. The method involves decomposing full questions into smaller segments, generating fabricated answers, and obfuscating the model's memory of the original input. A benchmark dataset was crafted with questions containing privacy-sensitive information from diverse fields. P2F achieves zero-shot generalization, allowing adaptability across a wide range of use cases without manual adjustments. Experimental results indicate P2F's robust capability to obfuscate LLM's memory, attaining a forgetfulness score of around 90\% without any utility loss. This represents an enhancement of up to 63\% when contrasted with the naive direct instruction technique, highlighting P2F's efficacy in mitigating memory retention of sensitive information within LLMs. Our findings establish the first benchmark in the novel field of the LLM forgetting task, representing a meaningful advancement in privacy preservation in the emerging LLM domain.
Modern healthcare often utilises radiographic images alongside textual reports for diagnostics, encouraging the use of Vision-Language Self-Supervised Learning (VL-SSL) with large pre-trained models to learn versatile medical vision representations. However, most existing VL-SSL frameworks are trained end-to-end, which is computation-heavy and can lose vital prior information embedded in pre-trained encoders. To address both issues, we introduce the backbone-agnostic Adaptor framework, which preserves medical knowledge in pre-trained image and text encoders by keeping them frozen, and employs a lightweight Adaptor module for cross-modal learning. Experiments on medical image classification and segmentation tasks across three datasets reveal that our framework delivers competitive performance while cutting trainable parameters by over 90% compared to current pre-training approaches. Notably, when fine-tuned with just 1% of data, Adaptor outperforms several Transformer-based methods trained on full datasets in medical image segmentation.
The review summarizes the main methodological concepts used in studying natural language from the perspective of complexity science and documents their applicability in identifying both universal and system-specific features of language in its written representation. Three main complexity-related research trends in quantitative linguistics are covered. The first part addresses the issue of word frequencies in texts and demonstrates that taking punctuation into consideration restores scaling whose violation in the Zipf's law is often observed for the most frequent words. The second part introduces methods inspired by time series analysis, used in studying various kinds of correlations in written texts. The related time series are generated on the basis of text partition into sentences or into phrases between consecutive punctuation marks. It turns out that these series develop features often found in signals generated by complex systems, like long-range correlations or (multi)fractal structures. Moreover, it appears that the distances between punctuation marks comply with the discrete variant of the Weibull distribution. In the third part, the application of the network formalism to natural language is reviewed, particularly in the context of the so-called word-adjacency networks. Parameters characterizing topology of such networks can be used for classification of texts, for example, from a stylometric perspective. Network approach can also be applied to represent the organization of word associations. Structure of word-association networks turns out to be significantly different from that observed in random networks, revealing genuine properties of language. Finally, punctuation seems to have a significant impact not only on the language's information-carrying ability but also on its key statistical properties, hence it is recommended to consider punctuation marks on a par with words.
As the presence of flying robots continues to grow in both commercial and private sectors, it necessitates an understanding of appropriate methods for nonverbal interaction with humans. While visual cues, such as gestures incorporated into trajectories, are more apparent and thoroughly researched, acoustic cues have remained unexplored, despite their potential to enhance human-drone interaction. Given that additional audiovisual and sensory equipment is not always desired or practicable, and flight noise often masks potential acoustic communication in rotary-wing drones, such as through a loudspeaker, the rotors themselves offer potential for nonverbal communication. In this paper, quadrotor trajectories are augmented by acoustic information that does not visually affect the flight, but adds audible information that significantly facilitates distinctiveness. A user study (N=192) demonstrates that sonically augmenting the trajectories of two aerial gestures makes them more easily distinguishable. This enhancement contributes to human-drone interaction through onboard means, particularly in situations where the human cannot see or look at the drone.
Large Language Models (LLMs) demonstrate significant capabilities but face challenges such as hallucination, outdated knowledge, and non-transparent, untraceable reasoning processes. Retrieval-Augmented Generation (RAG) has emerged as a promising solution by incorporating knowledge from external databases. This enhances the accuracy and credibility of the models, particularly for knowledge-intensive tasks, and allows for continuous knowledge updates and integration of domain-specific information. RAG synergistically merges LLMs' intrinsic knowledge with the vast, dynamic repositories of external databases. This comprehensive review paper offers a detailed examination of the progression of RAG paradigms, encompassing the Naive RAG, the Advanced RAG, and the Modular RAG. It meticulously scrutinizes the tripartite foundation of RAG frameworks, which includes the retrieval , the generation and the augmentation techniques. The paper highlights the state-of-the-art technologies embedded in each of these critical components, providing a profound understanding of the advancements in RAG systems. Furthermore, this paper introduces the metrics and benchmarks for assessing RAG models, along with the most up-to-date evaluation framework. In conclusion, the paper delineates prospective avenues for research, including the identification of challenges, the expansion of multi-modalities, and the progression of the RAG infrastructure and its ecosystem.
Energy disaggregation is a promising solution to access detailed information on energy consumption in a household, by itemizing its total energy consumption. However, in real-world applications, overfitting remains a challenging problem for data-driven disaggregation methods. First, the available real-world datasets are biased towards the most frequently used appliances. Second, both real and synthetic publicly-available datasets are limited in number of appliances, which may not be sufficient for a disaggregation algorithm to learn complex relations among different types of appliances and their states. To address the lack of appliance data, we propose two physics-informed data generators: one for high sampling rate signals (kHz) and another for low sampling rate signals (Hz). These generators rely on prior knowledge of the physics of appliance energy consumption, and are capable of simulating a virtually unlimited number of different appliances and their corresponding signatures for any time period. Both methods involve defining a mathematical model, selecting centroids corresponding to individual appliances, sampling model parameters around each centroid, and finally substituting the obtained parameters into the mathematical model. Additionally, by using Principal Component Analysis and Kullback-Leibler divergence, we demonstrate that our methods significantly outperform the previous approaches.
Visual scenes are extremely diverse, not only because there are infinite possible combinations of objects and backgrounds but also because the observations of the same scene may vary greatly with the change of viewpoints. When observing a multi-object visual scene from multiple viewpoints, humans can perceive the scene compositionally from each viewpoint while achieving the so-called ``object constancy'' across different viewpoints, even though the exact viewpoints are untold. This ability is essential for humans to identify the same object while moving and to learn from vision efficiently. It is intriguing to design models that have a similar ability. In this paper, we consider a novel problem of learning compositional scene representations from multiple unspecified (i.e., unknown and unrelated) viewpoints without using any supervision and propose a deep generative model which separates latent representations into a viewpoint-independent part and a viewpoint-dependent part to solve this problem. During the inference, latent representations are randomly initialized and iteratively updated by integrating the information in different viewpoints with neural networks. Experiments on several specifically designed synthetic datasets have shown that the proposed method can effectively learn from multiple unspecified viewpoints.
Striking a balance between precision and efficiency presents a prominent challenge in the bird's-eye-view (BEV) 3D object detection. Although previous camera-based BEV methods achieved remarkable performance by incorporating long-term temporal information, most of them still face the problem of low efficiency. One potential solution is knowledge distillation. Existing distillation methods only focus on reconstructing spatial features, while overlooking temporal knowledge. To this end, we propose TempDistiller, a Temporal knowledge Distiller, to acquire long-term memory from a teacher detector when provided with a limited number of frames. Specifically, a reconstruction target is formulated by integrating long-term temporal knowledge through self-attention operation applied to feature teachers. Subsequently, novel features are generated for masked student features via a generator. Ultimately, we utilize this reconstruction target to reconstruct the student features. In addition, we also explore temporal relational knowledge when inputting full frames for the student model. We verify the effectiveness of the proposed method on the nuScenes benchmark. The experimental results show our method obtain an enhancement of +1.6 mAP and +1.1 NDS compared to the baseline, a speed improvement of approximately 6 FPS after compressing temporal knowledge, and the most accurate velocity estimation.
This paper addresses the task of Unmanned Aerial Vehicles (UAV) visual geo-localization, which aims to match images of the same geographic target taken by different platforms, i.e., UAVs and satellites. In general, the key to achieving accurate UAV-satellite image matching lies in extracting visual features that are robust against viewpoint changes, scale variations, and rotations. Current works have shown that part matching is crucial for UAV visual geo-localization since part-level representations can capture image details and help to understand the semantic information of scenes. However, the importance of preserving semantic characteristics in part-level representations is not well discussed. In this paper, we introduce a transformer-based adaptive semantic aggregation method that regards parts as the most representative semantics in an image. Correlations of image patches to different parts are learned in terms of the transformer's feature map. Then our method decomposes part-level features into an adaptive sum of all patch features. By doing this, the learned parts are encouraged to focus on patches with typical semantics. Extensive experiments on the University-1652 dataset have shown the superiority of our method over the current works.
The values of two-player general-sum differential games are viscosity solutions to Hamilton-Jacobi-Isaacs (HJI) equations. Value and policy approximations for such games suffer from the curse of dimensionality (CoD). Alleviating CoD through physics-informed neural networks (PINN) encounters convergence issues when value discontinuity is present due to state constraints. On top of these challenges, it is often necessary to learn generalizable values and policies across a parametric space of games, e.g., for game parameter inference when information is incomplete. To address these challenges, we propose in this paper a Pontryagin-mode neural operator that outperforms existing state-of-the-art (SOTA) on safety performance across games with parametric state constraints. Our key contribution is the introduction of a costate loss defined on the discrepancy between forward and backward costate rollouts, which are computationally cheap. We show that the discontinuity of costate dynamics (in the presence of state constraints) effectively enables the learning of discontinuous values, without requiring manually supervised data as suggested by the current SOTA. More importantly, we show that the close relationship between costates and policies makes the former critical in learning feedback control policies with generalizable safety performance.