Search for: All records

Large language models (LLMs) have achieved remarkable performance on various natural language tasks. However, they are trained on static corpora and their knowledge can become outdated quickly in the fast-changing world. This motivates the development of knowledge editing (KE) to update specific knowledge in LLMs without changing unrelated others or compromising their pre-trained capabilities. Previous efforts sought to update a small amount of parameters of a LLM and proved effective for making selective updates. Nonetheless, the edited LLM often exhibits degraded ability to reason about the new knowledge. In this work, we identify a key issue: \textit{heterogeneous token overfitting} (HTO), where the LLM overfits different tokens in the provided knowledge at varying rates. To tackle this, we propose {\NAME}, a token-level smoothing method that mitigates HTO by adaptively refining the target distribution. Theoretically, {\NAME} offers better parameter updates with negligible computation overhead. It also induces an implicit DPO but does not require preference data pairs. Extensive experiments across four editing methods, two LLMs, and diverse scenarios demonstrate the effectiveness and versatility of our method. 

Abstract Differentiable rendering of translucent objects with respect to their shapes has been a long‐standing problem. State‐of‐the‐art methods require detecting object silhouettes or specifying change rates inside translucent objects—both of which can be expensive for translucent objects with complex shapes. In this paper, we address this problem for translucent objects with no refractive or reflective boundaries. By reparameterizing interior components of differential path integrals, our new formulation does not require change rates to be specified in the interior of objects. Further, we introduce new Monte Carlo estimators based on this formulation that do not require explicit detection of object silhouettes. 

Electric Vehicle (EV) charging recommendation that both accommodates user preference and adapts to the ever-changing external environment arises as a cost-effective strategy to alleviate the range anxiety of private EV drivers. Previous studies focus on centralized strategies to achieve optimized resource allocation, particularly useful for privacy-indifferent taxi fleets and fixed-route public transits. However, private EV driver seeks a more personalized and resource-aware charging recommendation that is tailor-made to accommodate the user preference (when and where to charge) yet sufficiently adaptive to the spatiotemporal mismatch between charging supply and demand. Here we propose a novel Regularized Actor-Critic (RAC) charging recommendation approach that would allow each EV driver to strike an optimal balance between the user preference (historical charging pattern) and the external reward (driving distance and wait time). Experimental results on two real-world datasets demonstrate the unique features and superior performance of our approach to the competing methods. 

Image retrieval relies heavily on the quality of the data modeling and the distance measurement in the feature space. Building on the concept of image manifold, we first propose to represent the feature space of images, learned via neural networks, as a graph. Neighborhoods in the feature space are now defined by the geodesic distance between images, represented as graph vertices or manifold samples. When limited images are available, this manifold is sparsely sampled, making the geodesic computation and the corresponding retrieval harder. To address this, we augment the manifold samples with geometrically aligned text, thereby using a plethora of sentences to teach us about images. In addition to extensive results on standard datasets illustrating the power of text to help in image retrieval, a new public dataset based on CLEVR is introduced to quantify the semantic similarity between visual data and text data. The experimental results show that the joint embedding manifold is a robust representation, allowing it to be a better basis to perform image retrieval given only an image and a textual instruction on the desired modifications over the image. 

Abstract Mathematically representing the shape of an object is a key ingredient for solving inverse rendering problems. Explicit representations like meshes are efficient to render in a differentiable fashion but have difficulties handling topology changes. Implicit representations like signed‐distance functions, on the other hand, offer better support of topology changes but are much more difficult to use for physics‐based differentiable rendering. We introduce a new physics‐based inverse rendering pipeline that uses both implicit and explicit representations. Our technique enjoys the benefit of both representations by supporting both topology changes and differentiable rendering of complex effects such as environmental illumination, soft shadows, and interreflection. We demonstrate the effectiveness of our technique using several synthetic and real examples.