Yuhang Song

Tommaso Salvatori, Yuhang Song, Yordan Yordanov et al. · oxford

Predictive coding networks are neuroscience-inspired models with roots in both Bayesian statistics and neuroscience. Training such models, however, is quite inefficient and unstable. In this work, we show how by simply changing the temporal scheduling of the update rule for the synaptic weights leads to an algorithm that is much more efficient and stable than the original one, and has theoretical guarantees in terms of convergence. The proposed algorithm, that we call incremental predictive coding (iPC) is also more biologically plausible than the original one, as it it fully automatic. In an extensive set of experiments, we show that iPC constantly performs better than the original formulation on a large number of benchmarks for image classification, as well as for the training of both conditional and masked language models, in terms of test accuracy, efficiency, and convergence with respect to a large set of hyperparameters.

Luca Pinchetti, Tommaso Salvatori, Yordan Yordanov et al. · oxford

A large amount of recent research has the far-reaching goal of finding training methods for deep neural networks that can serve as alternatives to backpropagation (BP). A prominent example is predictive coding (PC), which is a neuroscience-inspired method that performs inference on hierarchical Gaussian generative models. These methods, however, fail to keep up with modern neural networks, as they are unable to replicate the dynamics of complex layers and activation functions. In this work, we solve this problem by generalizing PC to arbitrary probability distributions, enabling the training of architectures, such as transformers, that are hard to approximate with only Gaussian assumptions. We perform three experimental analyses. First, we study the gap between our method and the standard formulation of PC on multiple toy examples. Second, we test the reconstruction quality on variational autoencoders, where our method reaches the same reconstruction quality as BP. Third, we show that our method allows us to train transformer networks and achieve a performance comparable with BP on conditional language models. More broadly, this method allows neuroscience-inspired learning to be applied to multiple domains, since the internal distributions can be flexibly adapted to the data, tasks, and architectures used.

Beren Millidge, Yuhang Song, Tommaso Salvatori et al.

How the brain performs credit assignment is a fundamental unsolved problem in neuroscience. Many `biologically plausible' algorithms have been proposed, which compute gradients that approximate those computed by backpropagation (BP), and which operate in ways that more closely satisfy the constraints imposed by neural circuitry. Many such algorithms utilize the framework of energy-based models (EBMs), in which all free variables in the model are optimized to minimize a global energy function. However, in the literature, these algorithms exist in isolation and no unified theory exists linking them together. Here, we provide a comprehensive theory of the conditions under which EBMs can approximate BP, which lets us unify many of the BP approximation results in the literature (namely, predictive coding, equilibrium propagation, and contrastive Hebbian learning) and demonstrate that their approximation to BP arises from a simple and general mathematical property of EBMs at free-phase equilibrium. This property can then be exploited in different ways with different energy functions, and these specific choices yield a family of BP-approximating algorithms, which both includes the known results in the literature and can be used to derive new ones.

Beren Millidge, Yuhang Song, Tommaso Salvatori et al.

Predictive coding (PC) is an influential theory in computational neuroscience, which argues that the cortex forms unsupervised world models by implementing a hierarchical process of prediction error minimization. PC networks (PCNs) are trained in two phases. First, neural activities are updated to optimize the network's response to external stimuli. Second, synaptic weights are updated to consolidate this change in activity -- an algorithm called \emph{prospective configuration}. While previous work has shown how in various limits, PCNs can be found to approximate backpropagation (BP), recent work has demonstrated that PCNs operating in this standard regime, which does not approximate BP, nevertheless obtain competitive training and generalization performance to BP-trained networks while outperforming them on tasks such as online, few-shot, and continual learning, where brains are known to excel. Despite this promising empirical performance, little is understood theoretically about the properties and dynamics of PCNs in this regime. In this paper, we provide a comprehensive theoretical analysis of the properties of PCNs trained with prospective configuration. We first derive analytical results concerning the inference equilibrium for PCNs and a previously unknown close connection relationship to target propagation (TP). Secondly, we provide a theoretical analysis of learning in PCNs as a variant of generalized expectation-maximization and use that to prove the convergence of PCNs to critical points of the BP loss function, thus showing that deep PCNs can, in theory, achieve the same generalization performance as BP, while maintaining their unique advantages.

Lei Sha, Yuhang Song, Yordan Yordanov et al. · oxford

Transformers have become an indispensable module for text generation models since their great success in machine translation. Previous works attribute the~success of transformers to the query-key-value dot-product attention, which provides a robust inductive bias by the fully connected token graphs. However, we found that self-attention has a severe limitation. When predicting the (i+1)-th token, self-attention only takes the i-th token as an information collector, and it tends to give a high attention weight to those tokens similar to itself. Therefore, most of the historical information that occurred before the i-th token is not taken into consideration. Based on this observation, in this paper, we propose a new architecture, called bird-eye transformer(BET), which goes one step further to improve the performance of transformers by reweighting self-attention to encourage it to focus more on important historical information. We have conducted experiments on multiple text generation tasks, including machine translation (2 datasets) and language models (3 datasets). These experimental~results show that our proposed model achieves a better performance than the baseline transformer architectures on~all~datasets. The code is released at: \url{https://sites.google.com/view/bet-transformer/home}.

Viet Dung Nguyen, Yuhang Song, Anh Nguyen et al.

Robot reinforcement learning from demonstrations (RLfD) assumes that expert data is abundant; this is usually unrealistic in the real world given data scarcity as well as high collection cost. Furthermore, imitation learning algorithms assume that the data is independently and identically distributed, which ultimately results in poorer performance as gradual errors emerge and compound within test-time trajectories. We address these issues by introducing the "master your own expertise" (MYOE) framework, a self-imitation framework that enables robotic agents to learn complex behaviors from limited demonstration data samples. Inspired by human perception and action, we propose and design what we call the queryable mixture-of-preferences state space model (QMoP-SSM), which estimates the desired goal at every time step. These desired goals are used in computing the "preference regret", which is used to optimize the robot control policy. Our experiments demonstrate the robustness, adaptability, and out-of-sample performance of our agent compared to other state-of-the-art RLfD schemes. The GitHub repository that supports this work can be found at: https://github.com/rxng8/neurorobot-preference-regret-learning.

Yuhang Song, Anh Nguyen, Chun-Yi Lee

This paper tackles the critical challenge of object navigation in autonomous navigation systems, particularly focusing on the problem of target approach and episode termination in environments with long optimal episode length in Deep Reinforcement Learning (DRL) based methods. While effective in environment exploration and object localization, conventional DRL methods often struggle with optimal path planning and termination recognition due to a lack of depth information. To overcome these limitations, we propose a novel approach, namely the Depth-Inference Termination Agent (DITA), which incorporates a supervised model called the Judge Model to implicitly infer object-wise depth and decide termination jointly with reinforcement learning. We train our judge model along with reinforcement learning in parallel and supervise the former efficiently by reward signal. Our evaluation shows the method is demonstrating superior performance, we achieve a 9.3% gain on success rate than our baseline method across all room types and gain 51.2% improvements on long episodes environment while maintaining slightly better Success Weighted by Path Length (SPL). Code and resources, visualization are available at: https://github.com/HuskyKingdom/DITA_acml2023

Yuhang Song, Andrzej Wojcicki, Thomas Lukasiewicz et al.

Learning agents that are not only capable of taking tests, but also innovating is becoming a hot topic in AI. One of the most promising paths towards this vision is multi-agent learning, where agents act as the environment for each other, and improving each agent means proposing new problems for others. However, existing evaluation platforms are either not compatible with multi-agent settings, or limited to a specific game. That is, there is not yet a general evaluation platform for research on multi-agent intelligence. To this end, we introduce Arena, a general evaluation platform for multi-agent intelligence with 35 games of diverse logics and representations. Furthermore, multi-agent intelligence is still at the stage where many problems remain unexplored. Therefore, we provide a building toolkit for researchers to easily invent and build novel multi-agent problems from the provided game set based on a GUI-configurable social tree and five basic multi-agent reward schemes. Finally, we provide Python implementations of five state-of-the-art deep multi-agent reinforcement learning baselines. Along with the baseline implementations, we release a set of 100 best agents/teams that we can train with different training schemes for each game, as the base for evaluating agents with population performance. As such, the research community can perform comparisons under a stable and uniform standard. All the implementations and accompanied tutorials have been open-sourced for the community at https://sites.google.com/view/arena-unity/.

Ying Chen, Jiajing Xie, Yuxiang Lin et al.

Multi-modal learning that combines pathological images with genomic data has significantly enhanced the accuracy of survival prediction. Nevertheless, existing methods have not fully utilized the inherent hierarchical structure within both whole slide images (WSIs) and transcriptomic data, from which better intra-modal representations and inter-modal integration could be derived. Moreover, many existing studies attempt to improve multi-modal representations through attention mechanisms, which inevitably lead to high complexity when processing high-dimensional WSIs and transcriptomic data. Recently, a structured state space model named Mamba emerged as a promising approach for its superior performance in modeling long sequences with low complexity. In this study, we propose Mamba with multi-grained multi-modal interaction (SurvMamba) for survival prediction. SurvMamba is implemented with a Hierarchical Interaction Mamba (HIM) module that facilitates efficient intra-modal interactions at different granularities, thereby capturing more detailed local features as well as rich global representations. In addition, an Interaction Fusion Mamba (IFM) module is used for cascaded inter-modal interactive fusion, yielding more comprehensive features for survival prediction. Comprehensive evaluations on five TCGA datasets demonstrate that SurvMamba outperforms other existing methods in terms of performance and computational cost.

Yuhang Song, Mario Gianni, Chenguang Yang et al.

This paper addresses the challenge of fine-grained alignment in Vision-and-Language Navigation (VLN) tasks, where robots navigate realistic 3D environments based on natural language instructions. Current approaches use contrastive learning to align language with visual trajectory sequences. Nevertheless, they encounter difficulties with fine-grained vision negatives. To enhance cross-modal embeddings, we introduce a novel Bayesian Optimization-based adversarial optimization framework for creating fine-grained contrastive vision samples. To validate the proposed methodology, we conduct a series of experiments to assess the effectiveness of the enriched embeddings on fine-grained vision negatives. We conduct experiments on two common VLN benchmarks R2R and REVERIE, experiments on the them demonstrate that these embeddings benefit navigation, and can lead to a promising performance enhancement. Our source code and trained models are available at: https://anonymous.4open.science/r/FGVLN.

Tommaso Salvatori, Beren Millidge, Yuhang Song et al.

An autoassociative memory model is a function that, given a set of data points, takes as input an arbitrary vector and outputs the most similar data point from the memorized set. However, popular memory models fail to retrieve images even when the corruption is mild and easy to detect for a human evaluator. This is because similarities are evaluated in the raw pixel space, which does not contain any semantic information about the images. This problem can be easily solved by computing \emph{similarities} in an embedding space instead of the pixel space. We show that an effective way of computing such embeddings is via a network pretrained with a contrastive loss. As the dimension of embedding spaces is often significantly smaller than the pixel space, we also have a faster computation of similarity scores. We test this method on complex datasets such as CIFAR10 and STL10. An additional drawback of current models is the need of storing the whole dataset in the pixel space, which is often extremely large. We relax this condition and propose a class of memory models that only stores low-dimensional semantic embeddings, and uses them to retrieve similar, but not identical, memories. We demonstrate a proof of concept of this method on a simple task on the MNIST dataset.

Yuhang Song, Naima Abrar Shami, Romaric Duvignau et al.

As graphs scale to billions of nodes and edges, graph Machine Learning workloads are constrained by the cost of multi-hop traversals over exponentially growing neighborhoods. While various system-level and algorithmic optimizations have been proposed to accelerate Graph Neural Network (GNN) pipelines, data management and movement remain the primary bottlenecks at scale. In this paper, we explore whether graph sparsification, a well-established technique that reduces edges to create sparser neighborhoods, can serve as a lightweight pre-processing step to address these bottlenecks while preserving accuracy on node classification tasks. We develop an extensible experimental framework that enables systematic evaluation of how different sparsification methods affect the performance and accuracy of GNN models. We conduct the first comprehensive study of GNN training and inference on sparsified graphs, revealing several key findings. First, sparsification often preserves or even improves predictive performance. As an example, random sparsification raises the accuracy of the GAT model by 6.8% on the PubMed graph. Second, benefits increase with scale, substantially accelerating both training and inference. Our results show that the K-Neighbor sparsifier improves model serving performance on the Products graph by 11.7x with only a 0.7% accuracy drop. Importantly, we find that the computational overhead of sparsification is quickly amortized, making it practical for very large graphs.

LinFeng Li, Jian Zhao, Zepeng Yang et al.

We present a winning solution to RoboSense 2025 Track 4: Cross-Modal Drone Navigation. The task retrieves the most relevant geo-referenced image from a large multi-platform corpus (satellite/drone/ground) given a natural-language query. Two obstacles are severe inter-platform heterogeneity and a domain gap between generic training descriptions and platform-specific test queries. We mitigate these with a domain-aligned preprocessing pipeline and a Mixture-of-Experts (MoE) framework: (i) platform-wise partitioning, satellite augmentation, and removal of orientation words; (ii) an LLM-based caption refinement pipeline to align textual semantics with the distinct visual characteristics of each platform. Using BGE-M3 (text) and EVA-CLIP (image), we train three platform experts using a progressive two-stage, hard-negative mining strategy to enhance discriminative power, and fuse their scores at inference. The system tops the official leaderboard, demonstrating robust cross-modal geo-localization under heterogeneous viewpoints.

Zuntao Liu, Hao Zhuang, Junjie Jiang et al.

Event cameras have the potential to capture continuous motion information over time and space, making them well-suited for optical flow estimation. However, most existing learning-based methods for event-based optical flow adopt frame-based techniques, ignoring the spatio-temporal characteristics of events. Additionally, these methods assume linear motion between consecutive events within the loss time window, which increases optical flow errors in long-time sequences. In this work, we observe that rich spatio-temporal information and accurate nonlinear motion between events are crucial for event-based optical flow estimation. Therefore, we propose E-NMSTFlow, a novel unsupervised event-based optical flow network focusing on long-time sequences. We propose a Spatio-Temporal Motion Feature Aware (STMFA) module and an Adaptive Motion Feature Enhancement (AMFE) module, both of which utilize rich spatio-temporal information to learn spatio-temporal data associations. Meanwhile, we propose a nonlinear motion compensation loss that utilizes the accurate nonlinear motion between events to improve the unsupervised learning of our network. Extensive experiments demonstrate the effectiveness and superiority of our method. Remarkably, our method ranks first among unsupervised learning methods on the MVSEC and DSEC-Flow datasets. Our project page is available at https://wynelio.github.io/E-NMSTFlow.

Beren Millidge, Tommaso Salvatori, Yuhang Song et al.

The backpropagation of error algorithm used to train deep neural networks has been fundamental to the successes of deep learning. However, it requires sequential backward updates and non-local computations, which make it challenging to parallelize at scale and is unlike how learning works in the brain. Neuroscience-inspired learning algorithms, however, such as \emph{predictive coding}, which utilize local learning, have the potential to overcome these limitations and advance beyond current deep learning technologies. While predictive coding originated in theoretical neuroscience as a model of information processing in the cortex, recent work has developed the idea into a general-purpose algorithm able to train neural networks using only local computations. In this survey, we review works that have contributed to this perspective and demonstrate the close theoretical connections between predictive coding and backpropagation, as well as works that highlight the multiple advantages of using predictive coding models over backpropagation-trained neural networks. Specifically, we show the substantially greater flexibility of predictive coding networks against equivalent deep neural networks, which can function as classifiers, generators, and associative memories simultaneously, and can be defined on arbitrary graph topologies. Finally, we review direct benchmarks of predictive coding networks on machine learning classification tasks, as well as its close connections to control theory and applications in robotics.

Beren Millidge, Tommaso Salvatori, Yuhang Song et al.

A large number of neural network models of associative memory have been proposed in the literature. These include the classical Hopfield networks (HNs), sparse distributed memories (SDMs), and more recently the modern continuous Hopfield networks (MCHNs), which possesses close links with self-attention in machine learning. In this paper, we propose a general framework for understanding the operation of such memory networks as a sequence of three operations: similarity, separation, and projection. We derive all these memory models as instances of our general framework with differing similarity and separation functions. We extend the mathematical framework of Krotov et al (2020) to express general associative memory models using neural network dynamics with only second-order interactions between neurons, and derive a general energy function that is a Lyapunov function of the dynamics. Finally, using our framework, we empirically investigate the capacity of using different similarity functions for these associative memory models, beyond the dot product similarity measure, and demonstrate empirically that Euclidean or Manhattan distance similarity metrics perform substantially better in practice on many tasks, enabling a more robust retrieval and higher memory capacity than existing models.

Tommaso Salvatori, Luca Pinchetti, Beren Millidge et al.

Training with backpropagation (BP) in standard deep learning consists of two main steps: a forward pass that maps a data point to its prediction, and a backward pass that propagates the error of this prediction back through the network. This process is highly effective when the goal is to minimize a specific objective function. However, it does not allow training on networks with cyclic or backward connections. This is an obstacle to reaching brain-like capabilities, as the highly complex heterarchical structure of the neural connections in the neocortex are potentially fundamental for its effectiveness. In this paper, we show how predictive coding (PC), a theory of information processing in the cortex, can be used to perform inference and learning on arbitrary graph topologies. We experimentally show how this formulation, called PC graphs, can be used to flexibly perform different tasks with the same network by simply stimulating specific neurons, and investigate how the topology of the graph influences the final performance. We conclude by comparing against simple baselines trained~with~BP.

Tommaso Salvatori, Yuhang Song, Yujian Hong et al.

Associative memories in the brain receive and store patterns of activity registered by the sensory neurons, and are able to retrieve them when necessary. Due to their importance in human intelligence, computational models of associative memories have been developed for several decades now. They include autoassociative memories, which allow for storing data points and retrieving a stored data point $s$ when provided with a noisy or partial variant of $s$, and heteroassociative memories, able to store and recall multi-modal data. In this paper, we present a novel neural model for realizing associative memories, based on a hierarchical generative network that receives external stimuli via sensory neurons. This model is trained using predictive coding, an error-based learning algorithm inspired by information processing in the cortex. To test the capabilities of this model, we perform multiple retrieval experiments from both corrupted and incomplete data points. In an extensive comparison, we show that this new model outperforms in retrieval accuracy and robustness popular associative memory models, such as autoencoders trained via backpropagation, and modern Hopfield networks. In particular, in completing partial data points, our model achieves remarkable results on natural image datasets, such as ImageNet, with a surprisingly high accuracy, even when only a tiny fraction of pixels of the original images is presented. Furthermore, we show that this method is able to handle multi-modal data, retrieving images from descriptions, and vice versa. We conclude by discussing the possible impact of this work in the neuroscience community, by showing that our model provides a plausible framework to study learning and retrieval of memories in the brain, as it closely mimics the behavior of the hippocampus as a memory index and generative model.

Tommaso Salvatori, Yuhang Song, Thomas Lukasiewicz et al.

Deep learning has redefined the field of artificial intelligence (AI) thanks to the rise of artificial neural networks, which are architectures inspired by their neurological counterpart in the brain. Through the years, this dualism between AI and neuroscience has brought immense benefits to both fields, allowing neural networks to be used in dozens of applications. These networks use an efficient implementation of reverse differentiation, called backpropagation (BP). This algorithm, however, is often criticized for its biological implausibility (e.g., lack of local update rules for the parameters). Therefore, biologically plausible learning methods that rely on predictive coding (PC), a framework for describing information processing in the brain, are increasingly studied. Recent works prove that these methods can approximate BP up to a certain margin on multilayer perceptrons (MLPs), and asymptotically on any other complex model, and that zero-divergence inference learning (Z-IL), a variant of PC, is able to exactly implement BP on MLPs. However, the recent literature shows also that there is no biologically plausible method yet that can exactly replicate the weight update of BP on complex models. To fill this gap, in this paper, we generalize (PC and) Z-IL by directly defining them on computational graphs, and show that it can perform exact reverse differentiation. What results is the first biologically plausible algorithm that is equivalent to BP in the way of updating parameters on any neural network, providing a bridge between the interdisciplinary research of neuroscience and deep learning.

Tommaso Salvatori, Yuhang Song, Thomas Lukasiewicz et al.

Predictive coding networks (PCNs) are an influential model for information processing in the brain. They have appealing theoretical interpretations and offer a single mechanism that accounts for diverse perceptual phenomena of the brain. On the other hand, backpropagation (BP) is commonly regarded to be the most successful learning method in modern machine learning. Thus, it is exciting that recent work formulates inference learning (IL) that trains PCNs to approximate BP. However, there are several remaining critical issues: (i) IL is an approximation to BP with unrealistic/non-trivial requirements, (ii) IL approximates BP in single-step weight updates; whether it leads to the same point as BP after the weight updates are conducted for more steps is unknown, and (iii) IL is computationally significantly more costly than BP. To solve these issues, a variant of IL that is strictly equivalent to BP in fully connected networks has been proposed. In this work, we build on this result by showing that it also holds for more complex architectures, namely, convolutional neural networks and (many-to-one) recurrent neural networks. To our knowledge, we are the first to show that a biologically plausible algorithm is able to exactly replicate the accuracy of BP on such complex architectures, bridging the existing gap between IL and BP, and setting an unprecedented performance for PCNs, which can now be considered as efficient alternatives to BP.

Jianyi Wang, Xin Deng, Mai Xu et al.

The past few years have witnessed fast development in video quality enhancement via deep learning. Existing methods mainly focus on enhancing the objective quality of compressed video while ignoring its perceptual quality. In this paper, we focus on enhancing the perceptual quality of compressed video. Our main observation is that enhancing the perceptual quality mostly relies on recovering high-frequency sub-bands in wavelet domain. Accordingly, we propose a novel generative adversarial network (GAN) based on multi-level wavelet packet transform (WPT) to enhance the perceptual quality of compressed video, which is called multi-level wavelet-based GAN (MW-GAN). In MW-GAN, we first apply motion compensation with a pyramid architecture to obtain temporal information. Then, we propose a wavelet reconstruction network with wavelet-dense residual blocks (WDRB) to recover the high-frequency details. In addition, the adversarial loss of MW-GAN is added via WPT to further encourage high-frequency details recovery for video frames. Experimental results demonstrate the superiority of our method.

Hao Sun, Ziping Xu, Yuhang Song et al.

Policy gradient (PG) algorithms have been widely used in reinforcement learning (RL). However, PG algorithms rely on exploiting the value function being learned with the first-order update locally, which results in limited sample efficiency. In this work, we propose an alternative method called Zeroth-Order Supervised Policy Improvement (ZOSPI). ZOSPI exploits the estimated value function $Q$ globally while preserving the local exploitation of the PG methods based on zeroth-order policy optimization. This learning paradigm follows Q-learning but overcomes the difficulty of efficiently operating argmax in continuous action space. It finds max-valued action within a small number of samples. The policy learning of ZOSPI has two steps: First, it samples actions and evaluates those actions with a learned value estimator, and then it learns to perform the action with the highest value through supervised learning. We further demonstrate such a supervised learning framework can learn multi-modal policies. Experiments show that ZOSPI achieves competitive results on the continuous control benchmarks with a remarkable sample efficiency.

Yuhang Song, Wenbo Li, Lei Zhang et al.

We study in this paper the problem of novel human-object interaction (HOI) detection, aiming at improving the generalization ability of the model to unseen scenarios. The challenge mainly stems from the large compositional space of objects and predicates, which leads to the lack of sufficient training data for all the object-predicate combinations. As a result, most existing HOI methods heavily rely on object priors and can hardly generalize to unseen combinations. To tackle this problem, we propose a unified framework of adversarial domain generalization to learn object-invariant features for predicate prediction. To measure the performance improvement, we create a new split of the HICO-DET dataset, where the HOIs in the test set are all unseen triplet categories in the training set. Our experiments show that the proposed framework significantly increases the performance by up to 50% on the new split of HICO-DET dataset and up to 125% on the UnRel dataset for auxiliary evaluation in detecting novel HOIs.

Rong Zhang, Wei Li, Peng Wang et al.

Motivated by the need for photo-realistic simulation in autonomous driving, in this paper we present a video inpainting algorithm \emph{AutoRemover}, designed specifically for generating street-view videos without any moving objects. In our setup we have two challenges: the first is the shadow, shadows are usually unlabeled but tightly coupled with the moving objects. The second is the large ego-motion in the videos. To deal with shadows, we build up an autonomous driving shadow dataset and design a deep neural network to detect shadows automatically. To deal with large ego-motion, we take advantage of the multi-source data, in particular the 3D data, in autonomous driving. More specifically, the geometric relationship between frames is incorporated into an inpainting deep neural network to produce high-quality structurally consistent video output. Experiments show that our method outperforms other state-of-the-art (SOTA) object removal algorithms, reducing the RMSE by over $19\%$.

Yuhang Song, Jianyi Wang, Thomas Lukasiewicz et al.

Intrinsic rewards were introduced to simulate how human intelligence works; they are usually evaluated by intrinsically-motivated play, i.e., playing games without extrinsic rewards but evaluated with extrinsic rewards. However, none of the existing intrinsic reward approaches can achieve human-level performance under this very challenging setting of intrinsically-motivated play. In this work, we propose a novel megalomania-driven intrinsic reward (called mega-reward), which, to our knowledge, is the first approach that achieves human-level performance in intrinsically-motivated play. Intuitively, mega-reward comes from the observation that infants' intelligence develops when they try to gain more control on entities in an environment; therefore, mega-reward aims to maximize the control capabilities of agents on given entities in a given environment. To formalize mega-reward, a relational transition model is proposed to bridge the gaps between direct and latent control. Experimental studies show that mega-reward (i) can greatly outperform all state-of-the-art intrinsic reward approaches, (ii) generally achieves the same level of performance as Ex-PPO and professional human-level scores, and (iii) has also a superior performance when it is incorporated with extrinsic rewards.

Yuhang Song, Jianyi Wang, Thomas Lukasiewicz et al.

Hierarchical reinforcement learning (HRL) has recently shown promising advances on speeding up learning, improving the exploration, and discovering intertask transferable skills. Most recent works focus on HRL with two levels, i.e., a master policy manipulates subpolicies, which in turn manipulate primitive actions. However, HRL with multiple levels is usually needed in many real-world scenarios, whose ultimate goals are highly abstract, while their actions are very primitive. Therefore, in this paper, we propose a diversity-driven extensible HRL (DEHRL), where an extensible and scalable framework is built and learned levelwise to realize HRL with multiple levels. DEHRL follows a popular assumption: diverse subpolicies are useful, i.e., subpolicies are believed to be more useful if they are more diverse. However, existing implementations of this diversity assumption usually have their own drawbacks, which makes them inapplicable to HRL with multiple levels. Consequently, we further propose a novel diversity-driven solution to achieve this assumption in DEHRL. Experimental studies evaluate DEHRL with five baselines from four perspectives in two domains; the results show that DEHRL outperforms the state-of-the-art baselines in all four aspects.

Jiali Duan, Xiaoyuan Guo, Yuhang Song et al.

Previous methods have dealt with discrete manipulation of facial attributes such as smile, sad, angry, surprise etc, out of canonical expressions and they are not scalable, operating in single modality. In this paper, we propose a novel framework that supports continuous edits and multi-modality portrait manipulation using adversarial learning. Specifically, we adapt cycle-consistency into the conditional setting by leveraging additional facial landmarks information. This has two effects: first cycle mapping induces bidirectional manipulation and identity preserving; second pairing samples from different modalities can thus be utilized. To ensure high-quality synthesis, we adopt texture-loss that enforces texture consistency and multi-level adversarial supervision that facilitates gradient flow. Quantitative and qualitative experiments show the effectiveness of our framework in performing flexible and multi-modality portrait manipulation with photo-realistic effects.

Yuhang Song, Chao Yang, Yeji Shen et al.

In this paper, we focus on image inpainting task, aiming at recovering the missing area of an incomplete image given the context information. Recent development in deep generative models enables an efficient end-to-end framework for image synthesis and inpainting tasks, but existing methods based on generative models don't exploit the segmentation information to constrain the object shapes, which usually lead to blurry results on the boundary. To tackle this problem, we propose to introduce the semantic segmentation information, which disentangles the inter-class difference and intra-class variation for image inpainting. This leads to much clearer recovered boundary between semantically different regions and better texture within semantically consistent segments. Our model factorizes the image inpainting process into segmentation prediction (SP-Net) and segmentation guidance (SG-Net) as two steps, which predict the segmentation labels in the missing area first, and then generate segmentation guided inpainting results. Experiments on multiple public datasets show that our approach outperforms existing methods in optimizing the image inpainting quality, and the interactive segmentation guidance provides possibilities for multi-modal predictions of image inpainting.

Chao Yang, Yuhang Song, Xiaofeng Liu et al.

Recent advances in deep generative models have shown promising potential in image inpanting, which refers to the task of predicting missing pixel values of an incomplete image using the known context. However, existing methods can be slow or generate unsatisfying results with easily detectable flaws. In addition, there is often perceivable discontinuity near the holes and require further post-processing to blend the results. We present a new approach to address the difficulty of training a very deep generative model to synthesize high-quality photo-realistic inpainting. Our model uses conditional generative adversarial networks (conditional GANs) as the backbone, and we introduce a novel block-wise procedural training scheme to stabilize the training while we increase the network depth. We also propose a new strategy called adversarial loss annealing to reduce the artifacts. We further describe several losses specifically designed for inpainting and show their effectiveness. Extensive experiments and user-study show that our approach outperforms existing methods in several tasks such as inpainting, face completion and image harmonization. Finally, we show our framework can be easily used as a tool for interactive guided inpainting, demonstrating its practical value to solve common real-world challenges.

Yuhang Song, Chao Yang, Zhe Lin et al.

We study the task of image inpainting, which is to fill in the missing region of an incomplete image with plausible contents. To this end, we propose a learning-based approach to generate visually coherent completion given a high-resolution image with missing components. In order to overcome the difficulty to directly learn the distribution of high-dimensional image data, we divide the task into inference and translation as two separate steps and model each step with a deep neural network. We also use simple heuristics to guide the propagation of local textures from the boundary to the hole. We show that, by using such techniques, inpainting reduces to the problem of learning two image-feature translation functions in much smaller space and hence easier to train. We evaluate our method on several public datasets and show that we generate results of better visual quality than previous state-of-the-art methods.

Yuhang Song, Mai Xu, Jianyi Wang et al.

Panoramic video provides immersive and interactive experience by enabling humans to control the field of view (FoV) through head movement (HM). Thus, HM plays a key role in modeling human attention on panoramic video. This paper establishes a database collecting subjects' HM in panoramic video sequences. From this database, we find that the HM data are highly consistent across subjects. Furthermore, we find that deep reinforcement learning (DRL) can be applied to predict HM positions, via maximizing the reward of imitating human HM scanpaths through the agent's actions. Based on our findings, we propose a DRL-based HM prediction (DHP) approach with offline and online versions, called offline-DHP and online-DHP. In offline-DHP, multiple DRL workflows are run to determine potential HM positions at each panoramic frame. Then, a heat map of the potential HM positions, named the HM map, is generated as the output of offline-DHP. In online-DHP, the next HM position of one subject is estimated given the currently observed HM position, which is achieved by developing a DRL algorithm upon the learned offline-DHP model. Finally, the experiments validate that our approach is effective in both offline and online prediction of HM positions for panoramic video, and that the learned offline-DHP model can improve the performance of online-DHP.

Yuhang Song, Main Xu, Songyang Zhang et al.

Deep learning (DL) advances state-of-the-art reinforcement learning (RL), by incorporating deep neural networks in learning representations from the input to RL. However, the conventional deep neural network architecture is limited in learning representations for multi-task RL (MT-RL), as multiple tasks can refer to different kinds of representations. In this paper, we thus propose a novel deep neural network architecture, namely generalization tower network (GTN), which can achieve MT-RL within a single learned model. Specifically, the architecture of GTN is composed of both horizontal and vertical streams. In our GTN architecture, horizontal streams are used to learn representation shared in similar tasks. In contrast, the vertical streams are introduced to be more suitable for handling diverse tasks, which encodes hierarchical shared knowledge of these tasks. The effectiveness of the introduced vertical stream is validated by experimental results. Experimental results further verify that our GTN architecture is able to advance the state-of-the-art MT-RL, via being tested on 51 Atari games.

Yuhang Song, Mai Xu, Shengxi Li

In video coding, compressed videos with certain and constant quality can ensure quality of experience (QoE). To this end, we propose in this paper a novel PID-based quality control (PQC) method for video coding. Specifically, a formulation is modelled to control quality of video coding with two objectives: minimizing control error and quality fluctuation. Then, we apply the Laplace domain analysis to model the relationship between quantization parameter (QP) and control error in this formulation. Given the relationship between QP and control error, we propose a solution to the PQC formulation, such that videos can be compressed at certain and constant quality. Finally, experimental results show that our PQC method is effective in both control accuracy and quality fluctuation.

Yuhang Song, Christopher Grimm, Xianming Wang et al.

We examine the problem of learning mappings from state to state, suitable for use in a model-based reinforcement-learning setting, that simultaneously generalize to novel states and can capture stochastic transitions. We show that currently popular generative adversarial networks struggle to learn these stochastic transition models but a modification to their loss functions results in a powerful learning algorithm for this class of problems.

Christopher Grimm, Yuhang Song, Michael L. Littman

Generative adversarial networks (GANs) are an exciting alternative to algorithms for solving density estimation problems---using data to assess how likely samples are to be drawn from the same distribution. Instead of explicitly computing these probabilities, GANs learn a generator that can match the given probabilistic source. This paper looks particularly at this matching capability in the context of problems with one-dimensional outputs. We identify a class of function decompositions with properties that make them well suited to the critic role in a leading approach to GANs known as Wasserstein GANs. We show that Taylor and Fourier series decompositions belong to our class, provide examples of these critics outperforming standard GAN approaches, and suggest how they can be scaled to higher dimensional problems in the future.

Qin Huang, Chunyang Xia, Chihao Wu et al.

Recent development in fully convolutional neural network enables efficient end-to-end learning of semantic segmentation. Traditionally, the convolutional classifiers are taught to learn the representative semantic features of labeled semantic objects. In this work, we propose a reverse attention network (RAN) architecture that trains the network to capture the opposite concept (i.e., what are not associated with a target class) as well. The RAN is a three-branch network that performs the direct, reverse and reverse-attention learning processes simultaneously. Extensive experiments are conducted to show the effectiveness of the RAN in semantic segmentation. Being built upon the DeepLabv2-LargeFOV, the RAN achieves the state-of-the-art mIoU score (48.1%) for the challenging PASCAL-Context dataset. Significant performance improvements are also observed for the PASCAL-VOC, Person-Part, NYUDv2 and ADE20K datasets.

Qin Huang, Chunyang Xia, Wenchao Zheng et al.

Semantic segmentation is critical to image content understanding and object localization. Recent development in fully-convolutional neural network (FCN) has enabled accurate pixel-level labeling. One issue in previous works is that the FCN based method does not exploit the object boundary information to delineate segmentation details since the object boundary label is ignored in the network training. To tackle this problem, we introduce a double branch fully convolutional neural network, which separates the learning of the desirable semantic class labeling with mask-level object proposals guided by relabeled boundaries. This network, called object boundary guided FCN (OBG-FCN), is able to integrate the distinct properties of object shape and class features elegantly in a fully convolutional way with a designed masking architecture. We conduct experiments on the PASCAL VOC segmentation benchmark, and show that the end-to-end trainable OBG-FCN system offers great improvement in optimizing the target semantic segmentation quality.