When an image generation process is guided by both a text prompt and spatial cues, such as a set of bounding boxes, do these elements work in harmony, or does one dominate the other? Our analysis of a pretrained image diffusion model that integrates gated self-attention into the U-Net reveals that spatial grounding often outweighs textual grounding due to the sequential flow from gated self-attention to cross-attention. We demonstrate that such bias can be significantly mitigated without sacrificing accuracy in either grounding by simply rewiring the network architecture, changing from sequential to parallel for gated self-attention and cross-attention. This surprisingly simple yet effective solution does not require any fine-tuning of the network but significantly reduces the trade-off between the two groundings. Our experiments demonstrate significant improvements from the original GLIGEN to the rewired version in the trade-off between textual grounding and spatial grounding.
相關內容
Recent works demonstrate a remarkable ability to customize text-to-image diffusion models while only providing a few example images. What happens if you try to customize such models using multiple, fine-grained concepts in a sequential (i.e., continual) manner? In our work, we show that recent state-of-the-art customization of text-to-image models suffer from catastrophic forgetting when new concepts arrive sequentially. Specifically, when adding a new concept, the ability to generate high quality images of past, similar concepts degrade. To circumvent this forgetting, we propose a new method, C-LoRA, composed of a continually self-regularized low-rank adaptation in cross attention layers of the popular Stable Diffusion model. Furthermore, we use customization prompts which do not include the word of the customized object (i.e., "person" for a human face dataset) and are initialized as completely random embeddings. Importantly, our method induces only marginal additional parameter costs and requires no storage of user data for replay. We show that C-LoRA not only outperforms several baselines for our proposed setting of text-to-image continual customization, which we refer to as Continual Diffusion, but that we achieve a new state-of-the-art in the well-established rehearsal-free continual learning setting for image classification. The high achieving performance of C-LoRA in two separate domains positions it as a compelling solution for a wide range of applications, and we believe it has significant potential for practical impact. Project page: //jamessealesmith.github.io/continual-diffusion/
We introduce a new regression framework designed to deal with large-scale, complex data that lies around a low-dimensional manifold with noises. Our approach first constructs a graph representation, referred to as the skeleton, to capture the underlying geometric structure. We then define metrics on the skeleton graph and apply nonparametric regression techniques, along with feature transformations based on the graph, to estimate the regression function. We also discuss the limitations of some nonparametric regressors with respect to the general metric space such as the skeleton graph. The proposed regression framework suggests a novel way to deal with data with underlying geometric structures and provides additional advantages in handling the union of multiple manifolds, additive noises, and noisy observations. We provide statistical guarantees for the proposed method and demonstrate its effectiveness through simulations and real data examples.
Linear representation learning is widely studied due to its conceptual simplicity and empirical utility in tasks such as compression, classification, and feature extraction. Given a set of points $[\mathbf{x}_1, \mathbf{x}_2, \ldots, \mathbf{x}_n] = \mathbf{X} \in \mathbb{R}^{d \times n}$ and a vector $\mathbf{y} \in \mathbb{R}^d$, the goal is to find coefficients $\mathbf{w} \in \mathbb{R}^n$ so that $\mathbf{X} \mathbf{w} \approx \mathbf{y}$, subject to some desired structure on $\mathbf{w}$. In this work we seek $\mathbf{w}$ that forms a local reconstruction of $\mathbf{y}$ by solving a regularized least squares regression problem. We obtain local solutions through a locality function that promotes the use of columns of $\mathbf{X}$ that are close to $\mathbf{y}$ when used as a regularization term. We prove that, for all levels of regularization and under a mild condition that the columns of $\mathbf{X}$ have a unique Delaunay triangulation, the optimal coefficients' number of non-zero entries is upper bounded by $d+1$, thereby providing local sparse solutions when $d \ll n$. Under the same condition we also show that for any $\mathbf{y}$ contained in the convex hull of $\mathbf{X}$ there exists a regime of regularization parameter such that the optimal coefficients are supported on the vertices of the Delaunay simplex containing $\mathbf{y}$. This provides an interpretation of the sparsity as having structure obtained implicitly from the Delaunay triangulation of $\mathbf{X}$. We demonstrate that our locality regularized problem can be solved in comparable time to other methods that identify the containing Delaunay simplex.
As the advent of artificial general intelligence (AGI) progresses at a breathtaking pace, the application of large language models (LLMs) as AI Agents in robotics remains in its nascent stage. A significant concern that hampers the seamless integration of these AI Agents into robotics is the unpredictability of the content they generate, a phenomena known as ``hallucination''. Drawing inspiration from biological neural systems, we propose a novel, layered architecture for autonomous robotics, bridging AI agent intelligence and robot instinct. In this context, we define Robot Instinct as the innate or learned set of responses and priorities in an autonomous robotic system that ensures survival-essential tasks, such as safety assurance and obstacle avoidance, are carried out in a timely and effective manner. This paradigm harmoniously combines the intelligence of LLMs with the instinct of robotic behaviors, contributing to a more safe and versatile autonomous robotic system. As a case study, we illustrate this paradigm within the context of a mobile robot, demonstrating its potential to significantly enhance autonomous robotics and enabling a future where robots can operate independently and safely across diverse environments.
As large language models (LLMs) take on complex tasks, their inputs are supplemented with longer contexts that incorporate domain knowledge or user-specific information. Yet using long contexts poses a challenge for responsive LLM systems, as nothing can be generated until the whole context is processed by the LLM. . CacheGen is a fast context-loading module for LLM systems. First, CacheGen uses a custom tensor encoder, which embraces KV cache's distributional properties, to encode a KV cache into more compact bitstream representations with negligible encoding/decoding overhead. This reduces the bandwidth demand to fetch the KV cache. Second, to maintain low context-loading delay and high generation quality, CacheGen adapts the streaming strategies to cope with changes in available bandwidth. When available bandwidth drops, CacheGen may raise the compression level for a part of the context or choose to recompute its KV cache on the fly. We test CacheGen on four popular LLMs of various sizes and four datasets (662 contexts in total). Compared to the recent systems that reuse the KV cache, CacheGen reduces the KV cache size by 3.5-4.3x and the total delay in fetching and processing contexts by 3.2-3.7x while having negligible impact on the LLM response quality in accuracy or perplexity.
As an emerging computing paradigm, edge computing offers computing resources closer to the data sources, helping to improve the service quality of many real-time applications. A crucial problem is designing a rational pricing mechanism to maximize the revenue of the edge computing service provider (ECSP). However, prior works have considerable limitations: clients are static and are required to disclose their preferences, which is impractical in reality. However, previous works assume user privacy information to be known or consider the number of users in edge scenarios to be static. To address this issue, we propose a novel sequential computation offloading mechanism, where the ECSP posts prices of computing resources with different configurations to clients in turn. Clients independently choose which computing resources to purchase and how to offload based on their prices. Then Egret, a deep reinforcement learning-based approach that achieves maximum revenue, is proposed. Egret determines the optimal price and visiting orders online without considering clients' preferences. Experimental results show that the revenue of ECSP in Egret is only 1.29\% lower than Oracle and 23.43\% better than the state-of-the-art when the client arrives dynamically.
Multi-object tracking (MOT) methods have seen a significant boost in performance recently, due to strong interest from the research community and steadily improving object detection methods. The majority of tracking methods follow the tracking-by-detection (TBD) paradigm, blindly trust the incoming detections with no sense of their associated localization uncertainty. This lack of uncertainty awareness poses a problem in safety-critical tasks such as autonomous driving where passengers could be put at risk due to erroneous detections that have propagated to downstream tasks, including MOT. While there are existing works in probabilistic object detection that predict the localization uncertainty around the boxes, no work in 2D MOT for autonomous driving has studied whether these estimates are meaningful enough to be leveraged effectively in object tracking. We introduce UncertaintyTrack, a collection of extensions that can be applied to multiple TBD trackers to account for localization uncertainty estimates from probabilistic object detectors. Experiments on the Berkeley Deep Drive MOT dataset show that the combination of our method and informative uncertainty estimates reduces the number of ID switches by around 19\% and improves mMOTA by 2-3%. The source code is available at //github.com/TRAILab/UncertaintyTrack
Interpretability methods are developed to understand the working mechanisms of black-box models, which is crucial to their responsible deployment. Fulfilling this goal requires both that the explanations generated by these methods are correct and that people can easily and reliably understand them. While the former has been addressed in prior work, the latter is often overlooked, resulting in informal model understanding derived from a handful of local explanations. In this paper, we introduce explanation summary (ExSum), a mathematical framework for quantifying model understanding, and propose metrics for its quality assessment. On two domains, ExSum highlights various limitations in the current practice, helps develop accurate model understanding, and reveals easily overlooked properties of the model. We also connect understandability to other properties of explanations such as human alignment, robustness, and counterfactual minimality and plausibility.
Visual dialogue is a challenging task that needs to extract implicit information from both visual (image) and textual (dialogue history) contexts. Classical approaches pay more attention to the integration of the current question, vision knowledge and text knowledge, despising the heterogeneous semantic gaps between the cross-modal information. In the meantime, the concatenation operation has become de-facto standard to the cross-modal information fusion, which has a limited ability in information retrieval. In this paper, we propose a novel Knowledge-Bridge Graph Network (KBGN) model by using graph to bridge the cross-modal semantic relations between vision and text knowledge in fine granularity, as well as retrieving required knowledge via an adaptive information selection mode. Moreover, the reasoning clues for visual dialogue can be clearly drawn from intra-modal entities and inter-modal bridges. Experimental results on VisDial v1.0 and VisDial-Q datasets demonstrate that our model outperforms exiting models with state-of-the-art results.
Many tasks in natural language processing can be viewed as multi-label classification problems. However, most of the existing models are trained with the standard cross-entropy loss function and use a fixed prediction policy (e.g., a threshold of 0.5) for all the labels, which completely ignores the complexity and dependencies among different labels. In this paper, we propose a meta-learning method to capture these complex label dependencies. More specifically, our method utilizes a meta-learner to jointly learn the training policies and prediction policies for different labels. The training policies are then used to train the classifier with the cross-entropy loss function, and the prediction policies are further implemented for prediction. Experimental results on fine-grained entity typing and text classification demonstrate that our proposed method can obtain more accurate multi-label classification results.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191