Tong Zhang

KaShun Shum, Shizhe Diao, Tong Zhang

Chain-of-thought (CoT) advances the reasoning abilities of large language models (LLMs) and achieves superior performance in complex reasoning tasks. However, most CoT studies rely on carefully designed human-annotated rational chains to prompt LLMs, posing challenges for real-world applications where labeled data is available without rational chains. This paper proposes a new strategy, Automate-CoT (Automatic Prompt Augmentation and Selection with Chain-of-Thought), that can bypass human engineering of CoT by automatically augmenting rational chains from a small labeled dataset, and then pruning low-quality chains to construct a candidate pool of machine-generated rationale chains based on the labels. Finally, it selects the optimal combination of several rationale chains from the pool for CoT prompting by employing a variance-reduced policy gradient strategy to estimate the significance of each example. Automate-CoT enables a quick adaptation of the CoT technique to different tasks. Experimental results demonstrate the effectiveness of our method, where competitive results are achieved on arithmetic reasoning (+2.7%), commonsense reasoning (+3.4%), symbolic reasoning (+3.2%), and non-reasoning tasks (+2.5%). The code is available at https://github.com/SHUMKASHUN/Automate-CoT.

Rui Xie, Asad Ul Haq, Linsen Ma et al.

Recent studies have revealed that, during the inference on generative AI models such as transformer, the importance of different weights exhibits substantial context-dependent variations. This naturally manifests a promising potential of adaptively configuring weight quantization to improve the generative AI inference efficiency. Although configurable weight quantization can readily leverage the hardware support of variable-precision arithmetics in modern GPU and AI accelerators, little prior research has studied how one could exploit variable weight quantization to proportionally improve the AI model memory access speed and energy efficiency. Motivated by the rapidly maturing CXL ecosystem, this work develops a CXL-based design solution to fill this gap. The key is to allow CXL memory controllers play an active role in supporting and exploiting runtime configurable weight quantization. Using transformer as a representative generative AI model, we carried out experiments that well demonstrate the effectiveness of the proposed design solution.

Yifan Hao, Chenlu Ye, Chi Han et al.

Transformer based models have shown remarkable capabilities in sequence learning across a wide range of tasks, often performing well on specific task by leveraging input-output examples. Despite their empirical success, a comprehensive theoretical understanding of this phenomenon remains limited. In this work, we investigate the layerwise behavior of Transformers to uncover the mechanisms underlying their multi-task generalization ability. Taking explorations on a typical sequence model, i.e, Hidden Markov Models, which are fundamental to many language tasks, we observe that: first, lower layers of Transformers focus on extracting feature representations, primarily influenced by neighboring tokens; second, on the upper layers, features become decoupled, exhibiting a high degree of time disentanglement. Building on these empirical insights, we provide theoretical analysis for the expressiveness power of Transformers. Our explicit constructions align closely with empirical observations, providing theoretical support for the Transformer's effectiveness and efficiency on sequence learning across diverse tasks.

Haohan Weng, Danqing Huang, Yu Qiao et al.

Templates serve as a good starting point to implement a design (e.g., banner, slide) but it takes great effort from designers to manually create. In this paper, we present Desigen, an automatic template creation pipeline which generates background images as well as harmonious layout elements over the background. Different from natural images, a background image should preserve enough non-salient space for the overlaying layout elements. To equip existing advanced diffusion-based models with stronger spatial control, we propose two simple but effective techniques to constrain the saliency distribution and reduce the attention weight in desired regions during the background generation process. Then conditioned on the background, we synthesize the layout with a Transformer-based autoregressive generator. To achieve a more harmonious composition, we propose an iterative inference strategy to adjust the synthesized background and layout in multiple rounds. We constructed a design dataset with more than 40k advertisement banners to verify our approach. Extensive experiments demonstrate that the proposed pipeline generates high-quality templates comparable to human designers. More than a single-page design, we further show an application of presentation generation that outputs a set of theme-consistent slides. The data and code are available at https://whaohan.github.io/desigen.

Jibao Qiu, C. L. Philip Chen, Tong Zhang

Symbolic Music Emotion Recognition(SMER) is to predict music emotion from symbolic data, such as MIDI and MusicXML. Previous work mainly focused on learning better representation via (mask) language model pre-training but ignored the intrinsic structure of the music, which is extremely important to the emotional expression of music. In this paper, we present a simple multi-task framework for SMER, which incorporates the emotion recognition task with other emotion-related auxiliary tasks derived from the intrinsic structure of the music. The results show that our multi-task framework can be adapted to different models. Moreover, the labels of auxiliary tasks are easy to be obtained, which means our multi-task methods do not require manually annotated labels other than emotion. Conducting on two publicly available datasets (EMOPIA and VGMIDI), the experiments show that our methods perform better in SMER task. Specifically, accuracy has been increased by 4.17 absolute point to 67.58 in EMOPIA dataset, and 1.97 absolute point to 55.85 in VGMIDI dataset. Ablation studies also show the effectiveness of multi-task methods designed in this paper.

Mengyue Zha, SiuTim Wong, Mengqi Liu et al.

This paper shows that masked autoencoder with extrapolator (ExtraMAE) is a scalable self-supervised model for time series generation. ExtraMAE randomly masks some patches of the original time series and learns temporal dynamics by recovering the masked patches. Our approach has two core designs. First, ExtraMAE is self-supervised. Supervision allows ExtraMAE to effectively and efficiently capture the temporal dynamics of the original time series. Second, ExtraMAE proposes an extrapolator to disentangle two jobs of the decoder: recovering latent representations and mapping them back into the feature space. These unique designs enable ExtraMAE to consistently and significantly outperform state-of-the-art (SoTA) benchmarks in time series generation. The lightweight architecture also makes ExtraMAE fast and scalable. ExtraMAE shows outstanding behavior in various downstream tasks such as time series classification, prediction, and imputation. As a self-supervised generative model, ExtraMAE allows explicit management of the synthetic data. We hope this paper will usher in a new era of time series generation with self-supervised models.