When limited by their own morphologies, humans and some species of animals have the remarkable ability to use objects from the environment toward accomplishing otherwise impossible tasks. Robots might similarly unlock a range of additional capabilities through tool use. Recent techniques for jointly optimizing morphology and control via deep learning are effective at designing locomotion agents. But while outputting a single morphology makes sense for locomotion, manipulation involves a variety of strategies depending on the task goals at hand. A manipulation agent must be capable of rapidly prototyping specialized tools for different goals. Therefore, we propose learning a designer policy, rather than a single design. A designer policy is conditioned on task information and outputs a tool design that helps solve the task. A design-conditioned controller policy can then perform manipulation using these tools. In this work, we take a step towards this goal by introducing a reinforcement learning framework for jointly learning these policies. Through simulated manipulation tasks, we show that this framework is more sample efficient than prior methods in multi-goal or multi-variant settings, can perform zero-shot interpolation or fine-tuning to tackle previously unseen goals, and allows tradeoffs between the complexity of design and control policies under practical constraints. Finally, we deploy our learned policies onto a real robot. Please see our supplementary video and website at //robotic-tool-design.github.io/ for visualizations.
相關內容
As the number of large language models (LLMs) released to the public grows, there is a pressing need to understand the safety implications associated with these models learning from third-party custom finetuning data. We explore the behavior of LLMs finetuned on noisy custom data containing unsafe content, represented by datasets that contain biases, toxicity, and harmfulness, finding that while aligned LLMs can readily learn this unsafe content, they also tend to forget it more significantly than other examples when subsequently finetuned on safer content. Drawing inspiration from the discrepancies in forgetting, we introduce the "ForgetFilter" algorithm, which filters unsafe data based on how strong the model's forgetting signal is for that data. We demonstrate that the ForgetFilter algorithm ensures safety in customized finetuning without compromising downstream task performance, unlike sequential safety finetuning. ForgetFilter outperforms alternative strategies like replay and moral self-correction in curbing LLMs' ability to assimilate unsafe content during custom finetuning, e.g. 75% lower than not applying any safety measures and 62% lower than using self-correction in toxicity score.
One of the primary reasons behind the success of neural networks has been the emergence of an array of new, highly-successful optimizers, perhaps most importantly the Adam optimizer. It is wiedely used for training neural networks, yet notoriously hard to interpret. Lacking a clear physical intuition, Adam is difficult to generalize to manifolds. Some attempts have been made to directly apply parts of the Adam algorithm to manifolds or to find an underlying structure, but a full generalization has remained elusive. In this work a new approach is presented that leverages the special structure of the manifolds which are relevant for optimization of neural networks, such as the Stiefel manifold, the symplectic Stiefel manifold, the Grassmann manifold and the symplectic Grassmann manifold: all of these are homogeneous spaces and as such admit a global tangent space representation. This global tangent space representation is used to perform all of the steps in the Adam optimizer. The resulting algorithm is then applied to train a transformer for which orthogonality constraints are enforced up to machine precision and we observe significant speed-ups in the training process. Optimization of neural networks where they weights do not lie on a manifold is identified as a special case of the presented framkework. This allows for a flexible implementation in which the learning rate is adapted simultaneously for all parameters, irrespective of whether they are an element of a general manifold or a vector space.
Perching on the moving platforms is a promising solution to enhance the endurance and operational range of quadrotors, which could benefit the efficiency of a variety of air-ground cooperative tasks. To ensure robust perching, tracking with a steady relative state and reliable perception is a prerequisite. This paper presents an adaptive dynamic tracking and perching scheme for autonomous quadrotors to achieve tight integration with moving platforms. For reliable perception of dynamic targets, we introduce elastic visibility-aware planning to actively avoid occlusion and target loss. Additionally, we propose a flexible terminal adjustment method that adapts the changes in flight duration and the coupled terminal states, ensuring full-state synchronization with the time-varying perching surface at various angles. A relaxation strategy is developed by optimizing the tangential relative speed to address the dynamics and safety violations brought by hard boundary conditions. Moreover, we take SE(3) motion planning into account to ensure no collision between the quadrotor and the platform until the contact moment. Furthermore, we propose an efficient spatiotemporal trajectory optimization framework considering full state dynamics for tracking and perching. The proposed method is extensively tested through benchmark comparisons and ablation studies. To facilitate the application of academic research to industry and to validate the efficiency of our scheme under strictly limited computational resources, we deploy our system on a commercial drone (DJI-MAVIC3) with a full-size sport-utility vehicle (SUV). We conduct extensive real-world experiments, where the drone successfully tracks and perches at 30~km/h (8.3~m/s) on the top of the SUV, and at 3.5~m/s with 60{\deg} inclined into the trunk of the SUV.
Child pornography represents a severe form of exploitation and victimization of children, leaving the victims with emotional and physical trauma. In this study, we aim to analyze local patterns of child pornography consumption across 1341 French communes in 20 metropolitan regions of France using fine-grained mobile traffic data of Tor network-related web services. We estimate that approx. 0.08 % of Tor mobile download traffic observed in France is linked to the consumption of child sexual abuse materials by correlating it with local-level temporal porn consumption patterns. This compares to 0.19 % of what we conservatively estimate to be the share of child pornographic content in global Tor traffic. In line with existing literature on the link between sexual child abuse and the consumption of image-based content thereof, we observe a positive and statistically significant effect of our child pornography consumption estimates on the reported number of victims of sexual violence and vice versa, which validates our findings, after controlling for a set of spatial and non-spatial features including socio-demographic characteristics, voting behaviour, nearby points of interest and Google Trends queries. While this is a first, exploratory attempt to look at child pornography from a spatial epidemiological angle, we believe this research provides public health officials with valuable information to prioritize target areas for public awareness campaigns as another step to fulfil the global community's pledge to target 16.2 of the Sustainable Development Goals: "End abuse, exploitation, trafficking and all forms of violence and torture against children".
Modeling the trajectories of animals is challenging due to the complexity of their behaviors, the influence of unpredictable environmental factors, individual variability, and the lack of detailed data on their movements. Additionally, factors such as migration, hunting, reproduction, and social interactions add additional layers of complexity when attempting to accurately forecast their movements. In the literature, various models exits that aim to study animal telemetry, by modeling the velocity of the telemetry, the telemetry itself or both processes jointly through a Markovian process. In this work, we propose to model the velocity of each coordinate axis for animal telemetry data as a fractional Ornstein-Uhlenbeck (fOU) process. Then, the integral fOU process models position data in animal telemetry. Compared to traditional methods, the proposed model is flexible in modeling long-range memory. The Hurst parameter $H \in (0,1)$ is a crucial parameter in integral fOU process, as it determines the degree of dependence or long-range memory. The integral fOU process is nonstationary process. In addition, a higher Hurst parameter ($H > 0.5$) indicates a stronger memory, leading to trajectories with transient trends, while a lower Hurst parameter ($H < 0.5$) implies a weaker memory, resulting in trajectories with recurring trends. When H = 0.5, the process reduces to a standard integral Ornstein-Uhlenbeck process. We develop a fast simulation algorithm of telemetry trajectories using an approach via finite-dimensional distributions. We also develop a maximum likelihood method for parameter estimation and its performance is examined by simulation studies. Finally, we present a telemetry application of Fin Whales that disperse over the Gulf of Mexico.
News outlets are now more than ever incentivized to provide their audience with slanted news, while the intrinsic homophilic nature of online social media may exacerbate polarized opinions. Here, we propose a new dynamic latent space model for time-varying online audience-duplication networks, which exploits social media content to conduct inference on media bias and polarization of news outlets. Our model contributes to the literature in several directions: 1) we provide a model-embedded data-driven interpretation for the latent leaning of news outlets in terms of media bias; 2) we endow our model with Markov-switching dynamics to capture polarization regimes while maintaining a parsimonious specification; 3) we contribute to the literature on the statistical properties of latent space network models. The proposed model is applied to a set of data on the online activity of national and local news outlets from four European countries in the years 2015 and 2016. We find evidence of a strong positive correlation between our media slant measure and a well-grounded external source of media bias. In addition, we provide insight into the polarization regimes across the four countries considered.
Ensembles over neural network weights trained from different random initialization, known as deep ensembles, achieve state-of-the-art accuracy and calibration. The recently introduced batch ensembles provide a drop-in replacement that is more parameter efficient. In this paper, we design ensembles not only over weights, but over hyperparameters to improve the state of the art in both settings. For best performance independent of budget, we propose hyper-deep ensembles, a simple procedure that involves a random search over different hyperparameters, themselves stratified across multiple random initializations. Its strong performance highlights the benefit of combining models with both weight and hyperparameter diversity. We further propose a parameter efficient version, hyper-batch ensembles, which builds on the layer structure of batch ensembles and self-tuning networks. The computational and memory costs of our method are notably lower than typical ensembles. On image classification tasks, with MLP, LeNet, and Wide ResNet 28-10 architectures, our methodology improves upon both deep and batch ensembles.
Humans and animals have the ability to continually acquire, fine-tune, and transfer knowledge and skills throughout their lifespan. This ability, referred to as lifelong learning, is mediated by a rich set of neurocognitive mechanisms that together contribute to the development and specialization of our sensorimotor skills as well as to long-term memory consolidation and retrieval. Consequently, lifelong learning capabilities are crucial for autonomous agents interacting in the real world and processing continuous streams of information. However, lifelong learning remains a long-standing challenge for machine learning and neural network models since the continual acquisition of incrementally available information from non-stationary data distributions generally leads to catastrophic forgetting or interference. This limitation represents a major drawback for state-of-the-art deep neural network models that typically learn representations from stationary batches of training data, thus without accounting for situations in which information becomes incrementally available over time. In this review, we critically summarize the main challenges linked to lifelong learning for artificial learning systems and compare existing neural network approaches that alleviate, to different extents, catastrophic forgetting. We discuss well-established and emerging research motivated by lifelong learning factors in biological systems such as structural plasticity, memory replay, curriculum and transfer learning, intrinsic motivation, and multisensory integration.
While it is nearly effortless for humans to quickly assess the perceptual similarity between two images, the underlying processes are thought to be quite complex. Despite this, the most widely used perceptual metrics today, such as PSNR and SSIM, are simple, shallow functions, and fail to account for many nuances of human perception. Recently, the deep learning community has found that features of the VGG network trained on the ImageNet classification task has been remarkably useful as a training loss for image synthesis. But how perceptual are these so-called "perceptual losses"? What elements are critical for their success? To answer these questions, we introduce a new Full Reference Image Quality Assessment (FR-IQA) dataset of perceptual human judgments, orders of magnitude larger than previous datasets. We systematically evaluate deep features across different architectures and tasks and compare them with classic metrics. We find that deep features outperform all previous metrics by huge margins. More surprisingly, this result is not restricted to ImageNet-trained VGG features, but holds across different deep architectures and levels of supervision (supervised, self-supervised, or even unsupervised). Our results suggest that perceptual similarity is an emergent property shared across deep visual representations.
Detecting carried objects is one of the requirements for developing systems to reason about activities involving people and objects. We present an approach to detect carried objects from a single video frame with a novel method that incorporates features from multiple scales. Initially, a foreground mask in a video frame is segmented into multi-scale superpixels. Then the human-like regions in the segmented area are identified by matching a set of extracted features from superpixels against learned features in a codebook. A carried object probability map is generated using the complement of the matching probabilities of superpixels to human-like regions and background information. A group of superpixels with high carried object probability and strong edge support is then merged to obtain the shape of the carried object. We applied our method to two challenging datasets, and results show that our method is competitive with or better than the state-of-the-art.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191