A Study on Training and Developing Large Language Models for Behavior Tree Generation

This paper presents an innovative exploration of the application potential of large language models (LLM) in addressing the challenging task of automatically generating behavior trees (BTs) for complex tasks. The conventional manual BT generation method is inefficient and heavily reliant on domain expertise. On the other hand, existing automatic BT generation technologies encounter bottlenecks related to task complexity, model adaptability, and reliability. In order to overcome these challenges, we propose a novel methodology that leverages the robust representation and reasoning abilities of LLMs. The core contribution of this paper lies in the design of a BT generation framework based on LLM, which encompasses the entire process, from data synthesis and model training to application developing and data verification. Synthetic data is introduced to train the BT generation model (BTGen model), enhancing its understanding and adaptability to various complex tasks, thereby significantly improving its overall performance. In order to ensure the effectiveness and executability of the generated BTs, we emphasize the importance of data verification and introduce a multilevel verification strategy. Additionally, we explore a range of agent design and development schemes with LLM as the central element. We hope that the work in this paper may provide a reference for the researchers who are interested in BT generation based on LLMs.

Investigating the Use of Traveltime and Reflection Tomography for Deep Learning-Based Sound-Speed Estimation in Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) is actively being developed to quantify acoustic tissue properties such as the speed-of-sound (SOS). Although full-waveform inversion (FWI) is an effective method for accurate SOS reconstruction, it can be computationally challenging for large-scale problems. Deep learning-based image-to-image learned reconstruction (IILR) methods are being investigated as scalable and computationally efficient alternatives. This study investigates the impact of the chosen input modalities on IILR methods for high-resolution SOS reconstruction in USCT. The selected modalities are traveltime tomography (TT) and reflection tomography (RT), which produce a low-resolution SOS map and a reflectivity map, respectively. These modalities have been chosen for their lower computational cost relative to FWI and their capacity to provide complementary information: TT offers a direct -- while low resolution -- SOS measure, while RT reveals tissue boundary information. Systematic analyses were facilitated by employing a stylized USCT imaging system with anatomically realistic numerical breast phantoms. Within this testbed, a supervised convolutional neural network (CNN) was trained to map dual-channel (TT and RT images) to a high-resolution SOS map. Moreover, the CNN was fine-tuned using a weighted reconstruction loss that prioritized tumor regions to address tumor underrepresentation in the training dataset. To understand the benefits of employing dual-channel inputs, single-input CNNs were trained separately using inputs from each modality alone (TT or RT). The methods were assessed quantitatively using normalized root mean squared error and structural similarity index measure for reconstruction accuracy and receiver operating characteristic analysis to assess signal detection-based performance measures.

Retinex-guided Channel-grouping based Patch Swap for Arbitrary Style Transfer

The basic principle of the patch-matching based style transfer is to substitute the patches of the content image feature maps by the closest patches from the style image feature maps. Since the finite features harvested from one single aesthetic style image are inadequate to represent the rich textures of the content natural image, existing techniques treat the full-channel style feature patches as simple signal tensors and create new style feature patches via signal-level fusion, which ignore the implicit diversities existed in style features and thus fail for generating better stylised results. In this paper, we propose a Retinex theory guided, channel-grouping based patch swap technique to solve the above challenges. Channel-grouping strategy groups the style feature maps into surface and texture channels, which prevents the winner-takes-all problem. Retinex theory based decomposition controls a more stable channel code rate generation. In addition, we provide complementary fusion and multi-scale generation strategy to prevent unexpected black area and over-stylised results respectively. Experimental results demonstrate that the proposed method outperforms the existing techniques in providing more style-consistent textures while keeping the content fidelity.

VideoGen: A Reference-Guided Latent Diffusion Approach for High Definition Text-to-Video Generation

In this paper, we present VideoGen, a text-to-video generation approach, which can generate a high-definition video with high frame fidelity and strong temporal consistency using reference-guided latent diffusion. We leverage an off-the-shelf text-to-image generation model, e.g., Stable Diffusion, to generate an image with high content quality from the text prompt, as a reference image to guide video generation. Then, we introduce an efficient cascaded latent diffusion module conditioned on both the reference image and the text prompt, for generating latent video representations, followed by a flow-based temporal upsampling step to improve the temporal resolution. Finally, we map latent video representations into a high-definition video through an enhanced video decoder. During training, we use the first frame of a ground-truth video as the reference image for training the cascaded latent diffusion module. The main characterises of our approach include: the reference image generated by the text-to-image model improves the visual fidelity; using it as the condition makes the diffusion model focus more on learning the video dynamics; and the video decoder is trained over unlabeled video data, thus benefiting from high-quality easily-available videos. VideoGen sets a new state-of-the-art in text-to-video generation in terms of both qualitative and quantitative evaluation. See \url{https://videogen.github.io/VideoGen/} for more samples.

Learned Full Waveform Inversion Incorporating Task Information for Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) is an emerging imaging modality that holds great promise for breast imaging. Full-waveform inversion (FWI)-based image reconstruction methods incorporate accurate wave physics to produce high spatial resolution quantitative images of speed of sound or other acoustic properties of the breast tissues from USCT measurement data. However, the high computational cost of FWI reconstruction represents a significant burden for its widespread application in a clinical setting. The research reported here investigates the use of a convolutional neural network (CNN) to learn a mapping from USCT waveform data to speed of sound estimates. The CNN was trained using a supervised approach with a task-informed loss function aiming at preserving features of the image that are relevant to the detection of lesions. A large set of anatomically and physiologically realistic numerical breast phantoms (NBPs) and corresponding simulated USCT measurements was employed during training. Once trained, the CNN can perform real-time FWI image reconstruction from USCT waveform data. The performance of the proposed method was assessed and compared against FWI using a hold-out sample of 41 NBPs and corresponding USCT data. Accuracy was measured using relative mean square error (RMSE), structural self-similarity index measure (SSIM), and lesion detection performance (DICE score). This numerical experiment demonstrates that a supervised learning model can achieve accuracy comparable to FWI in terms of RMSE and SSIM, and better performance in terms of task performance, while significantly reducing computational time.

EEG-based Emotion Style Transfer Network for Cross-dataset Emotion Recognition

As the key to realizing aBCIs, EEG emotion recognition has been widely studied by many researchers. Previous methods have performed well for intra-subject EEG emotion recognition. However, the style mismatch between source domain (training data) and target domain (test data) EEG samples caused by huge inter-domain differences is still a critical problem for EEG emotion recognition. To solve the problem of cross-dataset EEG emotion recognition, in this paper, we propose an EEG-based Emotion Style Transfer Network (E2STN) to obtain EEG representations that contain the content information of source domain and the style information of target domain, which is called stylized emotional EEG representations. The representations are helpful for cross-dataset discriminative prediction. Concretely, E2STN consists of three modules, i.e., transfer module, transfer evaluation module, and discriminative prediction module. The transfer module encodes the domain-specific information of source and target domains and then re-constructs the source domain's emotional pattern and the target domain's statistical characteristics into the new stylized EEG representations. In this process, the transfer evaluation module is adopted to constrain the generated representations that can more precisely fuse two kinds of complementary information from source and target domains and avoid distorting. Finally, the generated stylized EEG representations are fed into the discriminative prediction module for final classification. Extensive experiments show that the E2STN can achieve the state-of-the-art performance on cross-dataset EEG emotion recognition tasks.

Vision Transformer with Attention Map Hallucination and FFN Compaction

Vision Transformer(ViT) is now dominating many vision tasks. The drawback of quadratic complexity of its token-wise multi-head self-attention (MHSA), is extensively addressed via either token sparsification or dimension reduction (in spatial or channel). However, the therein redundancy of MHSA is usually overlooked and so is the feed-forward network (FFN). To this end, we propose attention map hallucination and FFN compaction to fill in the blank. Specifically, we observe similar attention maps exist in vanilla ViT and propose to hallucinate half of the attention maps from the rest with much cheaper operations, which is called hallucinated-MHSA (hMHSA). As for FFN, we factorize its hidden-to-output projection matrix and leverage the re-parameterization technique to strengthen its capability, making it compact-FFN (cFFN). With our proposed modules, a 10$\%$-20$\%$ reduction of floating point operations (FLOPs) and parameters (Params) is achieved for various ViT-based backbones, including straight (DeiT), hybrid (NextViT) and hierarchical (PVT) structures, meanwhile, the performances are quite competitive.

Revisiting Neural Retrieval on Accelerators

Retrieval finds a small number of relevant candidates from a large corpus for information retrieval and recommendation applications. A key component of retrieval is to model (user, item) similarity, which is commonly represented as the dot product of two learned embeddings. This formulation permits efficient inference, commonly known as Maximum Inner Product Search (MIPS). Despite its popularity, dot products cannot capture complex user-item interactions, which are multifaceted and likely high rank. We hence examine non-dot-product retrieval settings on accelerators, and propose \textit{mixture of logits} (MoL), which models (user, item) similarity as an adaptive composition of elementary similarity functions. This new formulation is expressive, capable of modeling high rank (user, item) interactions, and further generalizes to the long tail. When combined with a hierarchical retrieval strategy, \textit{h-indexer}, we are able to scale up MoL to 100M corpus on a single GPU with latency comparable to MIPS baselines. On public datasets, our approach leads to uplifts of up to 77.3\% in hit rate (HR). Experiments on a large recommendation surface at Meta showed strong metric gains and reduced popularity bias, validating the proposed approach's performance and improved generalization.

Master: Meta Style Transformer for Controllable Zero-Shot and Few-Shot Artistic Style Transfer

Transformer-based models achieve favorable performance in artistic style transfer recently thanks to its global receptive field and powerful multi-head/layer attention operations. Nevertheless, the over-paramerized multi-layer structure increases parameters significantly and thus presents a heavy burden for training. Moreover, for the task of style transfer, vanilla Transformer that fuses content and style features by residual connections is prone to content-wise distortion. In this paper, we devise a novel Transformer model termed as \emph{Master} specifically for style transfer. On the one hand, in the proposed model, different Transformer layers share a common group of parameters, which (1) reduces the total number of parameters, (2) leads to more robust training convergence, and (3) is readily to control the degree of stylization via tuning the number of stacked layers freely during inference. On the other hand, different from the vanilla version, we adopt a learnable scaling operation on content features before content-style feature interaction, which better preserves the original similarity between a pair of content features while ensuring the stylization quality. We also propose a novel meta learning scheme for the proposed model so that it can not only work in the typical setting of arbitrary style transfer, but also adaptable to the few-shot setting, by only fine-tuning the Transformer encoder layer in the few-shot stage for one specific style. Text-guided few-shot style transfer is firstly achieved with the proposed framework. Extensive experiments demonstrate the superiority of Master under both zero-shot and few-shot style transfer settings.

DeltaEdit: Exploring Text-free Training for Text-Driven Image Manipulation

Text-driven image manipulation remains challenging in training or inference flexibility. Conditional generative models depend heavily on expensive annotated training data. Meanwhile, recent frameworks, which leverage pre-trained vision-language models, are limited by either per text-prompt optimization or inference-time hyper-parameters tuning. In this work, we propose a novel framework named \textit{DeltaEdit} to address these problems. Our key idea is to investigate and identify a space, namely delta image and text space that has well-aligned distribution between CLIP visual feature differences of two images and CLIP textual embedding differences of source and target texts. Based on the CLIP delta space, the DeltaEdit network is designed to map the CLIP visual features differences to the editing directions of StyleGAN at training phase. Then, in inference phase, DeltaEdit predicts the StyleGAN's editing directions from the differences of the CLIP textual features. In this way, DeltaEdit is trained in a text-free manner. Once trained, it can well generalize to various text prompts for zero-shot inference without bells and whistles. Code is available at https://github.com/Yueming6568/DeltaEdit.