Quantization stands as a pivotal technique for large language model (LLM) serving, yet it poses significant challenges particularly in achieving effective low-bit quantization. The limited numerical mapping makes the quantized model produce a non-trivial error, bringing out intolerable performance degration. This paper is anchored in the basic idea of model compression objectives, and delves into the layer-wise error distribution of LLMs during post-training quantization. Subsequently, we introduce ASER, an algorithm consisting of (1) Error Reconstruction: low-rank compensation for quantization error with LoRA-style matrices constructed by whitening SVD; (2) Activation Smoothing: outlier extraction to gain smooth activation and better error compensation. ASER is capable of quantizing typical LLMs to low-bit ones, particularly preserving accuracy even in W4A8 per-channel setup. Experimental results show that ASER is competitive among the state-of-the-art quantization algorithms, showing potential to activation quantization, with minor overhead.

Language models (LMs) only pretrained on a general and massive corpus usually cannot attain satisfying performance on domain-specific downstream tasks, and hence, applying domain-specific pretraining to LMs is a common and indispensable practice. However, domain-specific pretraining can be costly and time-consuming, hindering LMs' deployment in real-world applications. In this work, we consider the incapability to memorize domain-specific knowledge embedded in the general corpus with rare occurrences and long-tail distributions as the leading cause for pretrained LMs' inferior downstream performance. Analysis of Neural Tangent Kernels (NTKs) reveals that those long-tail data are commonly overlooked in the model's gradient updates and, consequently, are not effectively memorized, leading to poor domain-specific downstream performance. Based on the intuition that data with similar semantic meaning are closer in the embedding space, we devise a Cluster-guided Sparse Expert (CSE) layer to actively learn long-tail domain knowledge typically neglected in previous pretrained LMs. During pretraining, a CSE layer efficiently clusters domain knowledge together and assigns long-tail knowledge to designate extra experts. CSE is also a lightweight structure that only needs to be incorporated in several deep layers. With our training strategy, we found that during pretraining, data of long-tail knowledge gradually formulate isolated, outlier clusters in an LM's representation spaces, especially in deeper layers. Our experimental results show that only pretraining CSE-based LMs is enough to achieve superior performance than regularly pretrained-finetuned LMs on various downstream tasks, implying the prospects of domain-specific-pretraining-free language models.

Despite their prevalence in deep-learning communities, over-parameterized models convey high demands of computational costs for proper training. This work studies the fine-grained, modular-level learning dynamics of over-parameterized models to attain a more efficient and fruitful training strategy. Empirical evidence reveals that when scaling down into network modules, such as heads in self-attention models, we can observe varying learning patterns implicitly associated with each module's trainability. To describe such modular-level learning capabilities, we introduce a novel concept dubbed modular neural tangent kernel (mNTK), and we demonstrate that the quality of a module's learning is tightly associated with its mNTK's principal eigenvalue $\lambda_{\max}$. A large $\lambda_{\max}$ indicates that the module learns features with better convergence, while those miniature ones may impact generalization negatively. Inspired by the discovery, we propose a novel training strategy termed Modular Adaptive Training (MAT) to update those modules with their $\lambda_{\max}$ exceeding a dynamic threshold selectively, concentrating the model on learning common features and ignoring those inconsistent ones. Unlike most existing training schemes with a complete BP cycle across all network modules, MAT can significantly save computations by its partially-updating strategy and can further improve performance. Experiments show that MAT nearly halves the computational cost of model training and outperforms the accuracy of baselines.

Interior scene colorization is vastly demanded in areas such as personalized architecture design. Existing works either require manual efforts to colorize individual objects or conform to fixed color patterns automatically learned from prior knowledge, whilst neglecting user preference. Quantitatively identifying user preferences is challenging, particularly at the early stage of the design process. The 3D setup also presents new challenges as the inhabitant can observe from any possible viewpoint. We propose a representative view selection method based on visual attention and a progressive preference inference model. We particularly focus on the progressive integration of eye-tracked user preference, which enables the assistance in creativity support and allows the possibility of convergent thinking. A series of user studies have been conducted to validate the effectiveness of the proposed view selection method, preference inference model and the creativity support mechanism.

Color design for 3D indoor scenes is a challenging problem due to many factors that need to be balanced. Although learning from images is a commonly adopted strategy, this strategy may be more suitable for natural scenes in which objects tend to have relatively fixed colors. For interior scenes consisting mostly of man-made objects, creative yet reasonable color assignments are expected. We propose$C^{3}$C3Assignment, a system providing diverse suggestions for interior color design while satisfying general global and local rules including color compatibility, color mood, contrast, and user preference. We extend these constraints from the image domain to$\mathbb {R}^3$, and formulate 3D interior color design as an optimization problem. The design is accomplished in an omnidirectional manner to ensure a comfortable experience when the inhabitant observes the interior scene from possible positions and directions. We design a surrogate-assisted evolutionary algorithm to efficiently solve the highly nonlinear optimization problem for interactive applications, and investigate the system performance concerning problem complexity, solver convergence, and suggestion diversity. Preliminary user studies have been conducted to validate the rule extension from 2D to 3D and to verify system usability.