Abstract
We consider the problem of decision-making under uncertainty in an environment with safety constraints. Many business and industrial applications rely on real-time optimization with changing inputs to improve key performance indicators. In the case of unknown environmental characteristics, real-time optimization becomes challenging, particularly for the satisfaction of safety constraints. We propose the ARTEO algorithm, where we cast multi-armed bandits as a mathematical programming problem subject to safety constraints and learn the environmental characteristics through changes in optimization inputs and through exploration. We quantify the uncertainty in unknown characteristics by using Gaussian processes and incorporate it into the utility function as a contribution which drives exploration. We adaptively control the size of this contribution using a heuristic in accordance with the requirements of the environment. We guarantee the safety of our algorithm with a high probability through confidence bounds constructed under the regularity assumptions of Gaussian processes. Compared to existing safe-learning approaches, our algorithm does not require an exclusive exploration phase and follows the optimization goals even in the explored points, which makes it suitable for safety-critical systems. We demonstrate the safety and efficiency of our approach with two experiments: an industrial process and an online bid optimization benchmark problem.
Abstract
Ensuring the safety of question answering (QA) systems is critical for deploying them in biomedical and scientific domains. One approach to improving these systems uses natural language inference (NLI) to determine whether answers are supported, or entailed, by some background context. However, these systems are vulnerable to supporting an answer with a source that is wrong or misleading. Our work proposes a critical approach by selecting answers based on whether they have been contradicted by some background context. We evaluate this system on multiple choice and extractive QA and find that while the contradiction-based systems are competitive with and often better than entailment-only systems, models that incorporate contradiction, entailment, and QA model confidence scores together are the best. Based on this result, we explore unique opportunities for leveraging contradiction-based approaches such for improving interpretability and selecting better answers.
Keyword: scaling
Set based velocity shaping for robotic manipulators
Authors: Ryan McGovern, Nikolaos Athanasopolous, Seán McLoone
Subjects: Robotics (cs.RO); Systems and Control (eess.SY)
Abstract
We develop a new framework for trajectory planning on predefined paths, for general N-link manipulators. Different from previous approaches generating open-loop minimum time controllers or pre-tuned motion profiles by time-scaling, we establish analytic algorithms that recover all initial conditions that can be driven to the desirable target set while adhering to environment constraints. More technologically relevant, we characterise families of corresponding safe state-feedback controllers with several desirable properties. A key enabler in our framework is the introduction of a state feedback template, that induces ordering properties between trajectories of the resulting closed-loop system. The proposed structure allows working on the nonlinear system directly in both the analysis and synthesis problems. Both offline computations and online implementation are scalable with respect to the number of links of the manipulator. The results can potentially be used in a series of challenging problems: Numerical experiments on a commercial robotic manipulator demonstrate that efficient online implementation is possible.
Multi-Agent Contracts
Authors: Paul Duetting, Tomer Ezra, Michal Feldman, Thomas Kesselheim
Subjects: Computer Science and Game Theory (cs.GT)
Abstract
We study a natural combinatorial single-principal multi-agent contract design problem, in which a principal motivates a team of agents to exert effort toward a given task. At the heart of our model is a reward function, which maps the agent efforts to an expected reward of the principal. We seek to design computationally efficient algorithms for finding optimal (or near-optimal) linear contracts for reward functions that belong to the complement-free hierarchy. Our first main result gives constant-factor approximation algorithms for submodular and XOS reward functions, with value and demand oracles, respectively. It relies on an unconventional use of ``prices'' and (approximate) demand queries for selecting the set of agents that the principal should contract with, and exploits a novel scaling property of XOS functions and their marginals, which may be of independent interest. Our second main result is an $\Omega(\sqrt{n})$ impossibility for settings with $n$ agents and subadditive reward functions, even with demand oracle access. A striking feature of this impossibility is that it applies to subadditive functions that are constant-factor close to submodular. This presents a surprising departure from previous literature, e.g., on combinatorial auctions.
MGiaD: Multigrid in all dimensions. Efficiency and robustness by coarsening in resolution and channel dimensions
Authors: Antonia van Betteray, Matthias Rottmann, Karsten Kahl
Abstract
Current state-of-the-art deep neural networks for image classification are made up of 10 - 100 million learnable weights and are therefore inherently prone to overfitting. The complexity of the weight count can be seen as a function of the number of channels, the spatial extent of the input and the number of layers of the network. Due to the use of convolutional layers the scaling of weight complexity is usually linear with regards to the resolution dimensions, but remains quadratic with respect to the number of channels. Active research in recent years in terms of using multigrid inspired ideas in deep neural networks have shown that on one hand a significant number of weights can be saved by appropriate weight sharing and on the other that a hierarchical structure in the channel dimension can improve the weight complexity to linear. In this work, we combine these multigrid ideas to introduce a joint framework of multigrid inspired architectures, that exploit multigrid structures in all relevant dimensions to achieve linear weight complexity scaling and drastically reduced weight counts. Our experiments show that this structured reduction in weight count is able to reduce overfitting and thus shows improved performance over state-of-the-art ResNet architectures on typical image classification benchmarks at lower network complexity.
Massively Multilingual ASR on 70 Languages: Tokenization, Architecture, and Generalization Capabilities
Authors: Andros Tjandra, Nayan Singhal, David Zhang, Ozlem Kalinli, Abdelrahman Mohamed, Duc Le, Michael L. Seltzer
Subjects: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
Abstract
End-to-end multilingual ASR has become more appealing because of several reasons such as simplifying the training and deployment process and positive performance transfer from high-resource to low-resource languages. However, scaling up the number of languages, total hours, and number of unique tokens is not a trivial task. This paper explores large-scale multilingual ASR models on 70 languages. We inspect two architectures: (1) Shared embedding and output and (2) Multiple embedding and output model. In the shared model experiments, we show the importance of tokenization strategy across different languages. Later, we use our optimal tokenization strategy to train multiple embedding and output model to further improve our result. Our multilingual ASR achieves 13.9%-15.6% average WER relative improvement compared to monolingual models. We show that our multilingual ASR generalizes well on an unseen dataset and domain, achieving 9.5% and 7.5% WER on Multilingual Librispeech (MLS) with zero-shot and finetuning, respectively.
Keyword: calibration
ViSE: Vision-Based 3D Real-Time Shape Estimation of Continuously Deformable Robots
Authors: Hehui Zheng (1 and 2), Sebastian Pinzello (1), Barnabas Gavin Cangan (1), Thomas Buchner (1), Robert K. Katzschmann (1) ((1) Soft Robotics Lab ETH Zurich, (2) ETH AI Center)
Abstract
The precise control of soft and continuum robots requires knowledge of their shape. The shape of these robots has, in contrast to classical rigid robots, infinite degrees of freedom. To partially reconstruct the shape, proprioceptive techniques use built-in sensors resulting in inaccurate results and increased fabrication complexity. Exteroceptive methods so far rely on placing reflective markers on all tracked components and triangulating their position using multiple motion-tracking cameras. Tracking systems are expensive and infeasible for deformable robots interacting with the environment due to marker occlusion and damage. Here, we present a regression approach for 3D shape estimation using a convolutional neural network. The proposed approach takes advantage of data-driven supervised learning and is capable of real-time marker-less shape estimation during inference. Two images of a robotic system are taken simultaneously at 25 Hz from two different perspectives, and are fed to the network, which returns for each pair the parameterized shape. The proposed approach outperforms marker-less state-of-the-art methods by a maximum of 4.4\% in estimation accuracy while at the same time being more robust and requiring no prior knowledge of the shape. The approach can be easily implemented due to only requiring two color cameras without depth and not needing an explicit calibration of the extrinsic parameters. Evaluations on two types of soft robotic arms and a soft robotic fish demonstrate our method's accuracy and versatility on highly deformable systems in real-time. The robust performance of the approach against different scene modifications (camera alignment and brightness) suggests its generalizability to a wider range of experimental setups, which will benefit downstream tasks such as robotic grasping and manipulation.
Keyword: out of distribution detection
There is no result
Keyword: out-of-distribution detection
There is no result
Keyword: expected calibration error
There is no result
Keyword: overconfident
There is no result
Keyword: overconfidence
There is no result
Keyword: confidence
Adaptive Real Time Exploration and Optimization for Safety-Critical Systems
Using contradictions to improve QA systems
Keyword: scaling
Set based velocity shaping for robotic manipulators
Multi-Agent Contracts
MGiaD: Multigrid in all dimensions. Efficiency and robustness by coarsening in resolution and channel dimensions
Massively Multilingual ASR on 70 Languages: Tokenization, Architecture, and Generalization Capabilities
Keyword: calibration
ViSE: Vision-Based 3D Real-Time Shape Estimation of Continuously Deformable Robots