The complexity of designing reward functions has been a major obstacle to the wide application of deep reinforcement learning (RL) techniques. Describing an agent's desired behaviors and properties can be difficult, even for experts. A new paradigm called reinforcement learning from human preferences (or preference-based RL) has emerged as a promising solution, in which reward functions are learned from human preference labels among behavior trajectories. However, existing methods for preference-based RL are limited by the need for accurate oracle preference labels. This paper addresses this limitation by developing a method for crowd-sourcing preference labels and learning from diverse human preferences. The key idea is to stabilize reward learning through regularization and correction in a latent space. To ensure temporal consistency, a strong constraint is imposed on the reward model that forces its latent space to be close to the prior distribution. Additionally, a confidence-based reward model ensembling method is designed to generate more stable and reliable predictions. The proposed method is tested on a variety of tasks in DMcontrol and Meta-world and has shown consistent and significant improvements over existing preference-based RL algorithms when learning from diverse feedback, paving the way for real-world applications of RL methods.
Federated learning (FL) is a general principle for decentralized clients to train a server model collectively without sharing local data. FL is a promising framework with practical applications, but its standard training paradigm requires the clients to backpropagate through the model to compute gradients. Since these clients are typically edge devices and not fully trusted, executing backpropagation on them incurs computational and storage overhead as well as white-box vulnerability. In light of this, we develop backpropagation-free federated learning, dubbed BAFFLE, in which backpropagation is replaced by multiple forward processes to estimate gradients. BAFFLE is 1) memory-efficient and easily fits uploading bandwidth; 2) compatible with inference-only hardware optimization and model quantization or pruning; and 3) well-suited to trusted execution environments, because the clients in BAFFLE only execute forward propagation and return a set of scalars to the server. Empirically we use BAFFLE to train deep models from scratch or to finetune pretrained models, achieving acceptable results. Code is available in https://github.com/FengHZ/BAFFLE.
Unsupervised domain adaption has been widely adopted in tasks with scarce annotated data. Unfortunately, mapping the target-domain distribution to the source-domain unconditionally may distort the essential structural information of the target-domain data, leading to inferior performance. To address this issue, we firstly propose to introduce active sample selection to assist domain adaptation regarding the semantic segmentation task. By innovatively adopting multiple anchors instead of a single centroid, both source and target domains can be better characterized as multimodal distributions, in which way more complementary and informative samples are selected from the target domain. With only a little workload to manually annotate these active samples, the distortion of the target-domain distribution can be effectively alleviated, achieving a large performance gain. In addition, a powerful semi-supervised domain adaptation strategy is proposed to alleviate the long-tail distribution problem and further improve the segmentation performance. Extensive experiments are conducted on public datasets, and the results demonstrate that the proposed approach outperforms state-of-the-art methods by large margins and achieves similar performance to the fully-supervised upperbound, i.e., 71.4% mIoU on GTA5 and 71.8% mIoU on SYNTHIA. The effectiveness of each component is also verified by thorough ablation studies.
Vision-Language Pre-Training (VLP) has shown promising capabilities to align image and text pairs, facilitating a broad variety of cross-modal learning tasks. However, we observe that VLP models often lack the visual grounding/localization capability which is critical for many downstream tasks such as visual reasoning. In this work, we propose a novel Position-guided Text Prompt (PTP) paradigm to enhance the visual grounding ability of cross-modal models trained with VLP. Specifically, in the VLP phase, PTP divides the image into $N\times N$ blocks, and identifies the objects in each block through the widely used object detector in VLP. It then reformulates the visual grounding task into a fill-in-the-blank problem given a PTP by encouraging the model to predict the objects in the given blocks or regress the blocks of a given object, e.g. filling `P" or ``O" in aPTP ``The block P has a O". This mechanism improves the visual grounding capability of VLP models and thus helps them better handle various downstream tasks. By introducing PTP into several state-of-the-art VLP frameworks, we observe consistently significant improvements across representative cross-modal learning model architectures and several benchmarks, e.g. zero-shot Flickr30K Retrieval (+4.8 in average recall@1) for ViLT \cite{vilt} baseline, and COCO Captioning (+5.3 in CIDEr) for SOTA BLIP \cite{blip} baseline. Moreover, PTP achieves comparable results with object-detector based methods, and much faster inference speed since PTP discards its object detector for inference while the later cannot. Our code and pre-trained weight will be released at \url{https://github.com/sail-sg/ptp}.
Low-light stereo image enhancement (LLSIE) is a relatively new task to enhance the quality of visually unpleasant stereo images captured in dark condition. However, current methods achieve inferior performance on detail recovery and illumination adjustment. We find it is because: 1) the insufficient single-scale inter-view interaction makes the cross-view cues unable to be fully exploited; 2) lacking long-range dependency leads to the inability to deal with the spatial long-range effects caused by illumination degradation. To alleviate such limitations, we propose a LLSIE model termed Decoupled Cross-scale Cross-view Interaction Network (DCI-Net). Specifically, we present a decoupled interaction module (DIM) that aims for sufficient dual-view information interaction. DIM decouples the dual-view information exchange into discovering multi-scale cross-view correlations and further exploring cross-scale information flow. Besides, we present a spatial-channel information mining block (SIMB) for intra-view feature extraction, and the benefits are twofold. One is the long-range dependency capture to build spatial long-range relationship, and the other is expanded channel information refinement that enhances information flow in channel dimension. Extensive experiments on Flickr1024, KITTI 2012, KITTI 2015 and Middlebury datasets show that our method obtains better illumination adjustment and detail recovery, and achieves SOTA performance compared to other related methods. Our codes, datasets and models will be publicly available.
This paper is about an extraordinary phenomenon. Suppose we don't use any low-light images as training data, can we enhance a low-light image by deep learning? Obviously, current methods cannot do this, since deep neural networks require to train their scads of parameters using copious amounts of training data, especially task-related data. In this paper, we show that in the context of fundamental deep learning, it is possible to enhance a low-light image without any task-related training data. Technically, we propose a new, magical, effective and efficient method, termed \underline{Noi}se \underline{SE}lf-\underline{R}egression (NoiSER), which learns a gray-world mapping from Gaussian distribution for low-light image enhancement (LLIE). Specifically, a self-regression model is built as a carrier to learn a gray-world mapping during training, which is performed by simply iteratively feeding random noise. During inference, a low-light image is directly fed into the learned mapping to yield a normal-light one. Extensive experiments show that our NoiSER is highly competitive to current task-related data based LLIE models in terms of quantitative and visual results, while outperforming them in terms of the number of parameters, training time and inference speed. With only about 1K parameters, NoiSER realizes about 1 minute for training and 1.2 ms for inference with 600$\times$400 resolution on RTX 2080 Ti. Besides, NoiSER has an inborn automated exposure suppression capability and can automatically adjust too bright or too dark, without additional manipulations.
Low-light stereo image enhancement (LLSIE) is a relatively new task to enhance the quality of visually unpleasant stereo images captured in dark conditions. So far, very few studies on deep LLSIE have been explored due to certain challenging issues, i.e., the task has not been well addressed, and current methods clearly suffer from two shortages: 1) insufficient cross-view interaction; 2) lacking long-range dependency for intra-view learning. In this paper, we therefore propose a novel LLSIE model, termed \underline{Suf}ficient C\underline{r}oss-View \underline{In}teraction Network (SufrinNet). To be specific, we present sufficient inter-view interaction module (SIIM) to enhance the information exchange across views. SIIM not only discovers the cross-view correlations at different scales, but also explores the cross-scale information interaction. Besides, we present a spatial-channel information mining block (SIMB) for intra-view feature extraction, and the benefits are twofold. One is the long-range dependency capture to build spatial long-range relationship, and the other is expanded channel information refinement that enhances information flow in channel dimension. Extensive experiments on Flickr1024, KITTI 2012, KITTI 2015 and Middlebury datasets show that our method obtains better illumination adjustment and detail recovery, and achieves SOTA performance compared to other related methods. Our codes, datasets and models will be publicly available.
While deep generative models have empowered music generation, it remains a challenging and under-explored problem to edit an existing musical piece at fine granularity. In this paper, we propose SDMuse, a unified Stochastic Differential Music editing and generation framework, which can not only compose a whole musical piece from scratch, but also modify existing musical pieces in many ways, such as combination, continuation, inpainting, and style transferring. The proposed SDMuse follows a two-stage pipeline to achieve music generation and editing on top of a hybrid representation including pianoroll and MIDI-event. In particular, SDMuse first generates/edits pianoroll by iteratively denoising through a stochastic differential equation (SDE) based on a diffusion model generative prior, and then refines the generated pianoroll and predicts MIDI-event tokens auto-regressively. We evaluate the generated music of our method on ailabs1k7 pop music dataset in terms of quality and controllability on various music editing and generation tasks. Experimental results demonstrate the effectiveness of our proposed stochastic differential music editing and generation process, as well as the hybrid representations.
MetaFormer, the abstracted architecture of Transformer, has been found to play a significant role in achieving competitive performance. In this paper, we further explore the capacity of MetaFormer, again, without focusing on token mixer design: we introduce several baseline models under MetaFormer using the most basic or common mixers, and summarize our observations as follows: (1) MetaFormer ensures solid lower bound of performance. By merely adopting identity mapping as the token mixer, the MetaFormer model, termed IdentityFormer, achieves >80% accuracy on ImageNet-1K. (2) MetaFormer works well with arbitrary token mixers. When specifying the token mixer as even a random matrix to mix tokens, the resulting model RandFormer yields an accuracy of >81%, outperforming IdentityFormer. Rest assured of MetaFormer's results when new token mixers are adopted. (3) MetaFormer effortlessly offers state-of-the-art results. With just conventional token mixers dated back five years ago, the models instantiated from MetaFormer already beat state of the art. (a) ConvFormer outperforms ConvNeXt. Taking the common depthwise separable convolutions as the token mixer, the model termed ConvFormer, which can be regarded as pure CNNs, outperforms the strong CNN model ConvNeXt. (b) CAFormer sets new record on ImageNet-1K. By simply applying depthwise separable convolutions as token mixer in the bottom stages and vanilla self-attention in the top stages, the resulting model CAFormer sets a new record on ImageNet-1K: it achieves an accuracy of 85.5% at 224x224 resolution, under normal supervised training without external data or distillation. In our expedition to probe MetaFormer, we also find that a new activation, StarReLU, reduces 71% FLOPs of activation compared with GELU yet achieves better performance. We expect StarReLU to find great potential in MetaFormer-like models alongside other neural networks.
Despite the recent advancement in multi-agent reinforcement learning (MARL), the MARL agents easily overfit the training environment and perform poorly in the evaluation scenarios where other agents behave differently. Obtaining generalizable policies for MARL agents is thus necessary but challenging mainly due to complex multi-agent interactions. In this work, we model the problem with Markov Games and propose a simple yet effective method, ranked policy memory (RPM), to collect diverse multi-agent trajectories for training MARL policies with good generalizability. The main idea of RPM is to maintain a look-up memory of policies. In particular, we try to acquire various levels of behaviors by saving policies via ranking the training episode return, i.e., the episode return of agents in the training environment; when an episode starts, the learning agent can then choose a policy from the RPM as the behavior policy. This innovative self-play training framework leverages agents' past policies and guarantees the diversity of multi-agent interaction in the training data. We implement RPM on top of MARL algorithms and conduct extensive experiments on Melting Pot. It has been demonstrated that RPM enables MARL agents to interact with unseen agents in multi-agent generalization evaluation scenarios and complete given tasks, and it significantly boosts the performance up to 402% on average.