17 research outputs found
Recommended from our members
Robust Task Specification for Learning Systems
No full textThis dissertation considers how to evaluate and improve the robustness of AI systems in situations that are systematically different to those encountered during training. Specifically, we focus on test-time robustness for two particular ways of specifying tasks, and two specific forms of generalization. The first part of this dissertation focuses on learning tasks from demonstrations with imitation, while the second focuses on specifying tasks for large language models using natural language instructions.In the first part, we specifically consider the combinatorial and in-distribution generalization of imitation learning. Our first contribution is a benchmark for how well learned policies can generalize along various axes. The benchmark allows us to manipulate these axes independently to determine invariances and equivariances the policy has. Using this benchmark, we show that some basic computer vision techniques (augmentation, egocentric views) improve imitative generalization, but more sophisticated representation learning techniques do not.In the second part, we consider instruction-following language models and adversarial robustness, where a user is actively trying to provoke errors from the model. Here we contribute a large dataset of prompt injection attacks obtained from an online game, which we distill into a benchmark for language model robustness. We also consider a second type of adversarial attack called a jailbreak, and show that existing evaluations are insufficient to gauge the actual misuse potential of jailbreaking techniques. Thus we propose a new benchmark that identifies effective jailbreaks while correctly disregarding ineffective ones.This dissertation proposes several evaluations for challenging problems where existing algorithms fail: imitation learning algorithms struggle to generalize when only few demonstrations are available, and representation learning is not an easy fix. Likewise, the safeguards around large language models are easy for an adversary to subvert. These negative results point toward ways that AI systems could be improved to be more robust in unexpected circumstances; we describe these opportunities for future work in Chapter 6
A survey of Negro business units and the employment opportunities afforded by such units in Houston, Texas, 1942
No full text
dalawhatyoumust: Kaaps, translingualism and linguistic citizenship in Cape Town, South Africa
No full textIn 2016 Wayde Van Niekerk, a South African athlete of mixed-race heritage won an Olympic gold medal. In South Africa, his win caused hashtags such as #proudlysouthafrican, #blackexcellence and #colouredexcellence to trend online. By and large, these hashtags index the ongoing competitive discourses regarding nationalism, race and culture in Cape Town (cf. Author, 2018). Amongst these hashtags, however, was #dalawhatyoumust, a Kaaps hashtag generally meaning to �do what needs to be done�. Unlike the aforementioned hashtags, this one seems to cross the linguistic and racial divide despite its strong associations with Coloured1 people on the Cape Flats. The seemingly effortless uptake of this hashtag by diverse South Africans suggest that it has somehow become unmoored of its ethnic and linguistic inception. We explore the use of this Kaaps hashtag as a form of translingual practice which is affect-laden and trans portable across and between diverse users online and which promotes a particular �cool Capetonian� culture. Analyzing select posts from the #dalawhatyoumust thread on Facebook, we provide a nuanced look at #dala whatyoumust as an uplifting genre which proleptically advises nameless viewers of the importance of selfactualization, determination and aspiration. Additionally, we include Goffman�s (1974) framing foundation to investigate how positivistic discourse has been rhizomatically taken up by a �realm� of implicit collective users online. This research interrogates long-held ideological boundaries between Kaaps and legitimized Standard Afrikaans and standard English. We conclude with a focus on Kaaps hashtags as semiotic acts of Linguistic Citizenship (cf. Williams and Stroud, 2013) which allows for the conjoining of Kaaps with diverse audiences, complex trajectories, and an assortment of accompanying semiotics. Following Stroud (2018:3) we argue that this Kaaps hashtag has become a form of languaging that facilitates ��the building of broad affinities of speakers that cut across�divisions and borders, and that negotiate co-existence/co-habitation outside of common ground in recognition of equivocation�. In South Africa, division was the order of the day and when we explore contemporary ordinary moments posted by heterogenous users using #dalawhatyoumust (henceforth #dwym) we aim to explore the ordinariness of languaging which brings people together despite their race, linguistic background, and ethnicity, that is to say an affinity of �cool Capetonian� style
Action Schema Networks: Generalised Policies With Deep Learning
No full textIn this paper, we introduce the Action Schema Network (ASNet): a neural network architecture for learning generalised policies for probabilistic planning problems. By mimicking the relational structure of planning problems, ASNets are able to adopt a weight sharing scheme which allows the network to be applied to any problem from a given planning domain. This allows the cost of training the network to be amortised over all problems in that domain. Further, we propose a training method which balances exploration and supervised training on small problems to produce a policy which remains robust when evaluated on larger problems. In experiments, we show that ASNet's learning capability allows it to significantly outperform traditional non-learning planners in several challenging domains
ASNets: Deep Learning for Generalised Planning
No full textIn this paper, we discuss the learning of generalised policies for probabilistic and classical planning problems using Action Schema Networks (ASNets). The ASNet is a neural network architecture that exploits the relational structure of (P)PDDL planning problems to learn a common set of weights that can be applied to any problem in a domain. By mimicking the actions chosen by a traditional, non-learning planner on a handful of small problems in a domain, ASNets are able to learn a generalised reactive policy that can quickly solve much larger instances from the domain. This work extends the ASNet architecture to make it more expressive, while still remaining invariant to a range of symmetries that exist in PPDDL problems. We also present a thorough experimental evaluation of ASNets, including a comparison with heuristic search planners on seven probabilistic and deterministic domains, an extended evaluation on over 18,000 Blocksworld instances, and an ablation study. Finally, we show that sparsity-inducing regularisation can produce ASNets that are compact enough for humans to understand, yielding insights into how the structure of ASNets allows them to generalise across a domain
Guiding Search with Generalized Policies for Probabilistic Planning
No full textWe examine techniques for combining generalized policies with search algorithms to exploit the strengths and overcome the weaknesses of each when solving probabilistic planning problems. The Action Schema Network (ASNet) is a recent contribution to planning that uses deep learning and neural networks to learn generalized policies for probabilistic planning problems. ASNets are well suited to problems where local knowledge of the environment can be exploited to improve performance, but may fail to generalize to problems they were not trained on. Monte-Carlo Tree Search (MCTS) is a forward-chaining state space search algorithm for optimal decision making which performs simulations to incrementally build a search tree and estimate the values of each state. Although MCTS can achieve state-of-the-art results when paired with domain-specific knowledge, without this knowledge, MCTS requires a large number of simulations in order to obtain reliable state-value estimates. By combining ASNets with MCTS, we are able to improve the capability of an ASNet to generalize beyond the distribution of problems it was trained on, as well as enhance the navigation of the search space by MCTS
Guiding Search with Generalized Policies for Probabilistic Planning
No full textWe examine techniques for combining generalized policies with search algorithms to exploit the strengths and overcome the weaknesses of each when solving probabilistic planning problems. The Action Schema Network (ASNet) is a recent contribution to planning that uses deep learning and neural networks to learn generalized policies for probabilistic planning problems. ASNets are well suited to problems where local knowledge of the environment can be exploited to improve performance, but may fail to generalize to problems they were not trained on. Monte-Carlo Tree Search (MCTS) is a forward-chaining state space search algorithm for optimal decision making which performs simulations to incrementally build a search tree and estimate the values of each state. Although MCTS can achieve state-of-the-art results when paired with domain-specific knowledge, without this knowledge, MCTS requires a large number of simulations in order to obtain reliable state-value estimates. By combining AS-Nets with MCTS, we are able to improve the capability of an ASNet to generalize beyond the distribution of problems it was trained on, as well as enhance the navigation of the search space by MCTS.Felipe Trevizan and Sylvie Thiebaux aresupported by ARC project DP180103446 “On-line planning for constrained autonomous agents in an uncertain world
