We consider a robust reinforcement learning problem, where a learning agent learns from a simulated training environment. To account for the model mis-specification between this training environment and the real environment due to lack of data, we adopt a formulation of Bayesian risk MDP (BRMDP) with infinite horizon, which uses Bayesian posterior to estimate the transition model and impose a risk functional to account for the model uncertainty. Observations from the real environment that is out of the agent's control arrive periodically and are utilized by the agent to update the Bayesian posterior to reduce model uncertainty. We theoretically demonstrate that BRMDP balances the trade-off between robustness and conservativeness, and we further develop a multi-stage Bayesian risk-averse Q-learning algorithm to solve BRMDP with streaming observations from real environment. The proposed algorithm learns a risk-averse yet optimal policy that depends on the availability of real-world observations. We provide a theoretical guarantee of strong convergence for the proposed algorithm.

Minimizing the difference of two submodular (DS) functions is a problem that naturally occurs in various machine learning problems. Although it is well known that a DS problem can be equivalently formulated as the minimization of the difference of two convex (DC) functions, existing algorithms do not fully exploit this connection. A classical algorithm for DC problems is called the DC algorithm (DCA). We introduce variants of DCA and its complete form (CDCA) that we apply to the DC program corresponding to DS minimization. We extend existing convergence properties of DCA, and connect them to convergence properties on the DS problem. Our results on DCA match the theoretical guarantees satisfied by existing DS algorithms, while providing a more complete characterization of convergence properties. In the case of CDCA, we obtain a stronger local minimality guarantee. Our numerical results show that our proposed algorithms outperform existing baselines on two applications: speech corpus selection and feature selection.

Large language models are trained in two stages: (1) unsupervised pretraining from raw text, to learn general-purpose representations, and (2) large scale instruction tuning and reinforcement learning, to better align to end tasks and user preferences. We measure the relative importance of these two stages by training LIMA, a 65B parameter LLaMa language model fine-tuned with the standard supervised loss on only 1,000 carefully curated prompts and responses, without any reinforcement learning or human preference modeling. LIMA demonstrates remarkably strong performance, learning to follow specific response formats from only a handful of examples in the training data, including complex queries that range from planning trip itineraries to speculating about alternate history. Moreover, the model tends to generalize well to unseen tasks that did not appear in the training data. In a controlled human study, responses from LIMA are either equivalent or strictly preferred to GPT-4 in 43% of cases; this statistic is as high as 58% when compared to Bard and 65% versus DaVinci003, which was trained with human feedback. Taken together, these results strongly suggest that almost all knowledge in large language models is learned during pretraining, and only limited instruction tuning data is necessary to teach models to produce high quality output.

Despite its successes, to date Artificial Intelligence (AI) is still characterized by a number of shortcomings with regards to different application domains and goals. These limitations are arguably both conceptual (e.g., related to underlying theoretical models, such as symbolic vs. connectionist), and operational (e.g., related to robustness and ability to generalize). Biologically inspired AI, and more specifically brain-inspired AI, promises to provide further biological aspects beyond those that are already traditionally included in AI, making it possible to assess and possibly overcome some of its present shortcomings. This article examines some conceptual, technical, and ethical issues raised by the development and use of brain-inspired AI. Against this background, the paper asks whether there is anything ethically unique about brain-inspired AI. The aim of the paper is to introduce a method that has a heuristic nature and that can be applied to identify and address the ethical issues arising from brain-inspired AI. The conclusion resulting from the application of this method is that, compared to traditional AI, brain-inspired AI raises new foundational ethical issues and some new practical ethical issues, and exacerbates some of the issues raised by traditional AI.

The proposed approach is to formalise the probabilistic puzzle in equational FOL. Two formalisations are needed: one theory for all models of the given puzzle, and a second theory for the favorable models. Then Mace4 - that computes all the interpretation models of a FOL theory - is called twice. First, it is asked to compute all the possible models M p .Second, the additional constraint is added, and Mace4 computes only favourabile models M f. Finally, the definition of probability is applied: the number of favorable models is divided by the number of possible models. The proposed approach equips students from the logic tribe to find the correct solution for puzzles from the probabilitistic tribe, by using their favourite instruments: modelling and formalisation. I have exemplified here five probabilistic puzzles and how they can be solved by translating the min FOL and then find the corresponding interpretation models. Mace4 was the tool of choice here. Ongoing work is investigating the limits of this method on various collections of probabilistic puzzles

Reweighted wake-sleep (RWS) is a machine learning method for performing Bayesian inference in a very general class of models. RWS draws $K$ samples from an underlying approximate posterior, then uses importance weighting to provide a better estimate of the true posterior. RWS then updates its approximate posterior towards the importance-weighted estimate of the true posterior. However, recent work [Chattergee and Diaconis, 2018] indicates that the number of samples required for effective importance weighting is exponential in the number of latent variables. Attaining such a large number of importance samples is intractable in all but the smallest models. Here, we develop massively parallel RWS, which circumvents this issue by drawing $K$ samples of all $n$ latent variables, and individually reasoning about all $K^n$ possible combinations of samples. While reasoning about $K^n$ combinations might seem intractable, the required computations can be performed in polynomial time by exploiting conditional independencies in the generative model. We show considerable improvements over standard "global" RWS, which draws $K$ samples from the full joint.

Traditional approaches to the design of multi-agent navigation algorithms consider the environment as a fixed constraint, despite the influence of spatial constraints on agents' performance. Yet hand-designing conducive environment layouts is inefficient and potentially expensive. The goal of this paper is to consider the environment as a decision variable in a system-level optimization problem, where both agent performance and environment cost are incorporated. Towards this end, we propose novel problems of unprioritized and prioritized environment optimization, where the former considers agents unbiasedly and the latter accounts for agent priorities. We show, through formal proofs, under which conditions the environment can change while guaranteeing completeness (i.e., all agents reach goals), and analyze the role of agent priorities in the environment optimization. We proceed to impose real-world constraints on the environment optimization and formulate it mathematically as a constrained stochastic optimization problem. Since the relation between agents, environment and performance is challenging to model, we leverage reinforcement learning to develop a model-free solution and a primal-dual mechanism to handle constraints. Distinct information processing architectures are integrated for various implementation scenarios, including online/offline optimization and discrete/continuous environment. Numerical results corroborate the theory and demonstrate the validity and adaptability of our approach.

Policy gradient algorithms are an important family of deep reinforcement learning techniques. Many past research endeavors focused on using the first-order policy gradient information to train policy networks. Different from these works, we conduct research in this paper driven by the believe that properly utilizing and controlling Hessian information associated with the policy gradient can noticeably improve the performance of policy gradient algorithms. One key Hessian information that attracted our attention is the Hessian trace, which gives the divergence of the policy gradient vector field in the Euclidean policy parametric space. We set the goal to generalize this Euclidean policy parametric space into a general Riemmanian manifold by introducing a metric tensor field $g_ab$ in the parametric space. This is achieved through newly developed mathematical tools, deep learning algorithms, and metric tensor deep neural networks (DNNs). Armed with these technical developments, we propose a new policy gradient algorithm that learns to minimize the absolute divergence in the Riemannian manifold as an important regularization mechanism, allowing the Riemannian manifold to smoothen its policy gradient vector field. The newly developed algorithm is experimentally studied on several benchmark reinforcement learning problems. Our experiments clearly show that the new metric tensor regularized algorithm can significantly outperform its counterpart that does not use our regularization technique. Additional experimental analysis further suggests that the trained metric tensor DNN and the corresponding metric tensor $g_{ab}$ can effectively reduce the absolute divergence towards zero in the Riemannian manifold.

Unprecedented lockdowns at the start of the COVID-19 pandemic have drastically changed the routines of millions of people, potentially impacting important health-related behaviors. In this study, we use YouTube videos embedded in tweets about diet, exercise and fitness posted before and during COVID-19 to investigate the influence of the pandemic lockdowns on diet and nutrition. In particular, we examine the nutritional profile of the foods mentioned in the transcript, description and title of each video in terms of six macronutrients (protein, energy, fat, sodium, sugar, and saturated fat). These macronutrient values were further linked to demographics to assess if there are specific effects on those potentially having insufficient access to healthy sources of food. Interrupted time series analysis revealed a considerable shift in the aggregated macronutrient scores before and during COVID-19. In particular, whereas areas with lower incomes showed decrease in energy, fat, and saturated fat, those with higher percentage of African Americans showed an elevation in sodium. Word2Vec word similarities and odds ratio analysis suggested a shift from popular diets and lifestyle bloggers before the lockdowns to the interest in a variety of healthy foods, communal sharing of quick and easy recipes, as well as a new emphasis on comfort foods. To the best of our knowledge, this work is novel in terms of linking attention signals in tweets, content of videos, their nutrients profile, and aggregate demographics of the users. The insights made possible by this combination of resources are important for monitoring the secondary health effects of social distancing, and informing social programs designed to alleviate these effects.

This article proposes a meta-learning method for estimating the conditional average treatment effect (CATE) from a few observational data. The proposed method learns how to estimate CATEs from multiple tasks and uses the knowledge for unseen tasks. In the proposed method, based on the meta-learner framework, we decompose the CATE estimation problem into sub-problems. For each sub-problem, we formulate our estimation models using neural networks with task-shared and task-specific parameters. With our formulation, we can obtain optimal task-specific parameters in a closed form that are differentiable with respect to task-shared parameters, making it possible to perform effective meta-learning. The task-shared parameters are trained such that the expected CATE estimation performance in few-shot settings is improved by minimizing the difference between a CATE estimated with a large amount of data and one estimated with just a few data. Our experimental results demonstrate that our method outperforms the existing meta-learning approaches and CATE estimation methods.