This paper considers the problem of learning temporal task specifications, e.g. automata and temporal logic, from expert demonstrations. Task specifications are a class of sparse memory augmented rewards with explicit support for temporal and Boolean composition. Three features make learning temporal task specifications difficult: (1) the (countably) infinite number of tasks under consideration; (2) an a-priori ignorance of what memory is needed to encode the task; and (3) the discrete solution space - typically addressed by (brute force) enumeration. To overcome these hurdles, we propose Demonstration Informed Specification Search (DISS): a family of algorithms requiring only black box access to a maximum entropy planner and a task sampler from labeled examples. DISS then works by alternating between conjecturing labeled examples to make the provided demonstrations less surprising and sampling tasks consistent with the conjectured labeled examples. We provide a concrete implementation of DISS in the context of tasks described by Deterministic Finite Automata, and show that DISS is able to efficiently identify tasks from only one or two expert demonstrations.
相關內容
Humans naturally exploit haptic feedback during contact-rich tasks like loading a dishwasher or stocking a bookshelf. Current robotic systems focus on avoiding unexpected contact, often relying on strategically placed environment sensors. Recently, contact-exploiting manipulation policies have been trained in simulation and deployed on real robots. However, they require some form of real-world adaptation to bridge the sim-to-real gap, which might not be feasible in all scenarios. In this paper we train a contact-exploiting manipulation policy in simulation for the contact-rich household task of loading plates into a slotted holder, which transfers without any fine-tuning to the real robot. We investigate various factors necessary for this zero-shot transfer, like time delay modeling, memory representation, and domain randomization. Our policy transfers with minimal sim-to-real gap and significantly outperforms heuristic and learnt baselines. It also generalizes to plates of different sizes and weights. Demonstration videos and code are available at //sites.google.com/view/compliant-object-insertion.
Quantum data access and quantum processing can make certain classically intractable learning tasks feasible. However, quantum capabilities will only be available to a select few in the near future. Thus, reliable schemes that allow classical clients to delegate learning to untrusted quantum servers are required to facilitate widespread access to quantum learning advantages. Building on a recently introduced framework of interactive proof systems for classical machine learning, we develop a framework for classical verification of quantum learning. We exhibit learning problems that a classical learner cannot efficiently solve on their own, but that they can efficiently and reliably solve when interacting with an untrusted quantum prover. Concretely, we consider the problems of agnostic learning parities and Fourier-sparse functions with respect to distributions with uniform input marginal. We propose a new quantum data access model that we call "mixture-of-superpositions" quantum examples, based on which we give efficient quantum learning algorithms for these tasks. Moreover, we prove that agnostic quantum parity and Fourier-sparse learning can be efficiently verified by a classical verifier with only random example or statistical query access. Finally, we showcase two general scenarios in learning and verification in which quantum mixture-of-superpositions examples do not lead to sample complexity improvements over classical data. Our results demonstrate that the potential power of quantum data for learning tasks, while not unlimited, can be utilized by classical agents through interaction with untrusted quantum entities.
We propose a novel $K$-nearest neighbor resampling procedure for estimating the performance of a policy from historical data containing realized episodes of a decision process generated under a different policy. We focus on feedback policies that depend deterministically on the current state in environments with continuous state-action spaces and system-inherent stochasticity effected by chosen actions. Such settings are common in a wide range of high-stake applications and are actively investigated in the context of stochastic control. Our procedure exploits that similar state/action pairs (in a metric sense) are associated with similar rewards and state transitions. This enables our resampling procedure to tackle the counterfactual estimation problem underlying off-policy evaluation (OPE) by simulating trajectories similarly to Monte Carlo methods. Compared to other OPE methods, our algorithm does not require optimization, can be efficiently implemented via tree-based nearest neighbor search and parallelization and does not explicitly assume a parametric model for the environment's dynamics. These properties make the proposed resampling algorithm particularly useful for stochastic control environments. We prove that our method is statistically consistent in estimating the performance of a policy in the OPE setting under weak assumptions and for data sets containing entire episodes rather than independent transitions. To establish the consistency, we generalize Stone's Theorem, a well-known result in nonparametric statistics on local averaging, to include episodic data and the counterfactual estimation underlying OPE. Numerical experiments demonstrate the effectiveness of the algorithm in a variety of stochastic control settings including a linear quadratic regulator, trade execution in limit order books and online stochastic bin packing.
In this work we explore recent advances in instruction-tuning language models on a range of open instruction-following datasets. Despite recent claims that open models can be on par with state-of-the-art proprietary models, these claims are often accompanied by limited evaluation, making it difficult to compare models across the board and determine the utility of various resources. We provide a large set of instruction-tuned models from 6.7B to 65B parameters in size, trained on 12 instruction datasets ranging from manually curated (e.g., OpenAssistant) to synthetic and distilled (e.g., Alpaca) and systematically evaluate them on their factual knowledge, reasoning, multilinguality, coding, and open-ended instruction following abilities through a collection of automatic, model-based, and human-based metrics. We further introduce T\"ulu, our best performing instruction-tuned model suite finetuned on a combination of high-quality open resources. Our experiments show that different instruction-tuning datasets can uncover or enhance specific skills, while no single dataset (or combination) provides the best performance across all evaluations. Interestingly, we find that model and human preference-based evaluations fail to reflect differences in model capabilities exposed by benchmark-based evaluations, suggesting the need for the type of systemic evaluation performed in this work. Our evaluations show that the best model in any given evaluation reaches on average 83% of ChatGPT performance, and 68% of GPT-4 performance, suggesting that further investment in building better base models and instruction-tuning data is required to close the gap. We release our instruction-tuned models, including a fully finetuned 65B T\"ulu, along with our code, data, and evaluation framework at //github.com/allenai/open-instruct to facilitate future research.
Specifications of complex, large scale, computer software and hardware systems can be radically simplified by using simple maps from input sequences to output values. These "state machine maps" provide an alternative representation of classical Moore type state machines. Composition of state machine maps corresponds to state machine products and can be used to specify essentially any type of interconnection as well as parallel and distributed computation. State machine maps can also specify abstract properties of systems and are significantly more concise and scalable than traditional representations of automata. Examples included here include specifications of producer/consumer software, network distributed consensus, real-time digital circuits, and operating system scheduling. The motivation for this work comes from experience designing and developing operating systems and real-time software where weak methods for understanding and exploring designs is a well known handicap. The methods introduced here are based on ordinary discrete mathematics, primitive recursive functions and deterministic state machines and are intended, initially, to aid the intuition and understanding of the system developers. Staying strictly within the boundaries of classical deterministic state machines anchors the methods to the algebraic structures of automata and semigroups, obviates any need for axiomatic deduction systems, "formal methods", or extensions to the model, and makes the specifications more faithful to engineering practice. While state machine maps are obvious representations of state machines, the techniques introduced here for defining and composing them are novel. To illustrate applications, the paper includes a fairly detailed specification and verification of the well-known "Paxos" distributed consensus algorithm plus a number of simpler examples including a digital PID controller.
Agents navigating in 3D environments require some form of memory, which should hold a compact and actionable representation of the history of observations useful for decision taking and planning. In most end-to-end learning approaches the representation is latent and usually does not have a clearly defined interpretation, whereas classical robotics addresses this with scene reconstruction resulting in some form of map, usually estimated with geometry and sensor models and/or learning. In this work we propose to learn an actionable representation of the scene independently of the targeted downstream task and without explicitly optimizing reconstruction. The learned representation is optimized by a blind auxiliary agent trained to navigate with it on multiple short sub episodes branching out from a waypoint and, most importantly, without any direct visual observation. We argue and show that the blindness property is important and forces the (trained) latent representation to be the only means for planning. With probing experiments we show that the learned representation optimizes navigability and not reconstruction. On downstream tasks we show that it is robust to changes in distribution, in particular the sim2real gap, which we evaluate with a real physical robot in a real office building, significantly improving performance.
U-statistics play central roles in many statistical learning tools but face the haunting issue of scalability. Significant efforts have been devoted into accelerating computation by U-statistic reduction. However, existing results almost exclusively focus on power analysis, while little work addresses risk control accuracy -- comparatively, the latter requires distinct and much more challenging techniques. In this paper, we establish the first statistical inference procedure with provably higher-order accurate risk control for incomplete U-statistics. The sharpness of our new result enables us to reveal how risk control accuracy also trades off with speed for the first time in literature, which complements the well-known variance-speed trade-off. Our proposed general framework converts the long-standing challenge of formulating accurate statistical inference procedures for many different designs into a surprisingly routine task. This paper covers non-degenerate and degenerate U-statistics, and network moments. We conducted comprehensive numerical studies and observed results that validate our theory's sharpness. Our method also demonstrates effectiveness on real-world data applications.
The ongoing deep learning revolution has allowed computers to outclass humans in various games and perceive features imperceptible to humans during classification tasks. Current machine learning techniques have clearly distinguished themselves in specialized tasks. However, we have yet to see robots capable of performing multiple tasks at an expert level. Most work in this field is focused on the development of more sophisticated learning algorithms for a robot's controller given a largely static and presupposed robotic design. By focusing on the development of robotic bodies, rather than neural controllers, I have discovered that robots can be designed such that they overcome many of the current pitfalls encountered by neural controllers in multitask settings. Through this discovery, I also present novel metrics to explicitly measure the learning ability of a robotic design and its resistance to common problems such as catastrophic interference. Traditionally, the physical robot design requires human engineers to plan every aspect of the system, which is expensive and often relies on human intuition. In contrast, within the field of evolutionary robotics, evolutionary algorithms are used to automatically create optimized designs, however, such designs are often still limited in their ability to perform in a multitask setting. The metrics created and presented here give a novel path to automated design that allow evolved robots to synergize with their controller to improve the computational efficiency of their learning while overcoming catastrophic interference. Overall, this dissertation intimates the ability to automatically design robots that are more general purpose than current robots and that can perform various tasks while requiring less computation.
Personalized treatment effect estimates are often of interest in high-stakes applications -- thus, before deploying a model estimating such effects in practice, one needs to be sure that the best candidate from the ever-growing machine learning toolbox for this task was chosen. Unfortunately, due to the absence of counterfactual information in practice, it is usually not possible to rely on standard validation metrics for doing so, leading to a well-known model selection dilemma in the treatment effect estimation literature. While some solutions have recently been investigated, systematic understanding of the strengths and weaknesses of different model selection criteria is still lacking. In this paper, instead of attempting to declare a global `winner', we therefore empirically investigate success- and failure modes of different selection criteria. We highlight that there is a complex interplay between selection strategies, candidate estimators and the data used for comparing them, and provide interesting insights into the relative (dis)advantages of different criteria alongside desiderata for the design of further illuminating empirical studies in this context.
The time and effort involved in hand-designing deep neural networks is immense. This has prompted the development of Neural Architecture Search (NAS) techniques to automate this design. However, NAS algorithms tend to be slow and expensive; they need to train vast numbers of candidate networks to inform the search process. This could be alleviated if we could partially predict a network's trained accuracy from its initial state. In this work, we examine the overlap of activations between datapoints in untrained networks and motivate how this can give a measure which is usefully indicative of a network's trained performance. We incorporate this measure into a simple algorithm that allows us to search for powerful networks without any training in a matter of seconds on a single GPU, and verify its effectiveness on NAS-Bench-101, NAS-Bench-201, NATS-Bench, and Network Design Spaces. Our approach can be readily combined with more expensive search methods; we examine a simple adaptation of regularised evolutionary search. Code for reproducing our experiments is available at //github.com/BayesWatch/nas-without-training.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191