Fine-tuning pretrained language models (PLMs) for downstream tasks is a large-scale optimization problem, in which the choice of the training algorithm critically determines how well the trained model can generalize to unseen test data, especially in the context of few-shot learning. To achieve good generalization performance and avoid overfitting, techniques such as data augmentation and pruning are often applied. However, adding these regularizations necessitates heavy tuning of the hyperparameters of optimization algorithms, such as the popular Adam optimizer. In this paper, we propose a two-stage fine-tuning method, PAC-tuning, to address this optimization challenge. First, based on PAC-Bayes training, PAC-tuning directly minimizes the PAC-Bayes generalization bound to learn proper parameter distribution. Second, PAC-tuning modifies the gradient by injecting noise with the variance learned in the first stage into the model parameters during training, resulting in a variant of perturbed gradient descent (PGD). In the past, the few-shot scenario posed difficulties for PAC-Bayes training because the PAC-Bayes bound, when applied to large models with limited training data, might not be stringent. Our experimental results across 5 GLUE benchmark tasks demonstrate that PAC-tuning successfully handles the challenges of fine-tuning tasks and outperforms strong baseline methods by a visible margin, further confirming the potential to apply PAC training for any other settings where the Adam optimizer is currently used for training.

When multiple self-adaptive systems share the same environment and have common goals, they may coordinate their adaptations at runtime to avoid conflicts and to satisfy their goals. There are two approaches to coordination. (1) Logically centralized, where a supervisor has complete control over the individual self-adaptive systems. Such approach is infeasible when the systems have different owners or administrative domains. (2) Logically decentralized, where coordination is achieved through direct interactions. Because the individual systems have control over the information they share, decentralized coordination accommodates multiple administrative domains. However, existing techniques do not account simultaneously for both local concerns, e.g., preferences, and shared concerns, e.g., conflicts, which may lead to goals not being achieved as expected. Our idea to address this shortcoming is to express both types of concerns within the same constraint optimization problem. We propose CoADAPT, a decentralized coordination technique introducing two types of constraints: preference constraints, expressing local concerns, and consistency constraints, expressing shared concerns. At runtime, the problem is solved in a decentralized way using distributed constraint optimization algorithms implemented by each self-adaptive system. As a first step in realizing CoADAPT, we focus in this work on the coordination of adaptation planning strategies, traditionally addressed only with centralized techniques. We show the feasibility of CoADAPT in an exemplar from cloud computing and analyze experimentally its scalability.

Line search (or backtracking) procedures have been widely employed into first-order methods for solving convex optimization problems, especially those with unknown problem parameters (e.g., Lipschitz constant). In this paper, we show that line search is superfluous in attaining the optimal rate of convergence for solving a convex optimization problem whose parameters are not given a priori. In particular, we present a novel accelerated gradient descent type algorithm called auto-conditioned fast gradient method (AC-FGM) that can achieve an optimal $\mathcal{O}(1/k^2)$ rate of convergence for smooth convex optimization without requiring the estimate of a global Lipschitz constant or the employment of line search procedures. We then extend AC-FGM to solve convex optimization problems with H\"{o}lder continuous gradients and show that it automatically achieves the optimal rates of convergence uniformly for all problem classes with the desired accuracy of the solution as the only input. Finally, we report some encouraging numerical results that demonstrate the advantages of AC-FGM over the previously developed parameter-free methods for convex optimization.

Counterfactual explanations play an important role in detecting bias and improving the explainability of data-driven classification models. A counterfactual explanation (CE) is a minimal perturbed data point for which the decision of the model changes. Most of the existing methods can only provide one CE, which may not be achievable for the user. In this work we derive an iterative method to calculate robust CEs, i.e. CEs that remain valid even after the features are slightly perturbed. To this end, our method provides a whole region of CEs allowing the user to choose a suitable recourse to obtain a desired outcome. We use algorithmic ideas from robust optimization and prove convergence results for the most common machine learning methods including logistic regression, decision trees, random forests, and neural networks. Our experiments show that our method can efficiently generate globally optimal robust CEs for a variety of common data sets and classification models.

We consider the problem of learning classification trees that are robust to distribution shifts between training and testing/deployment data. This problem arises frequently in high stakes settings such as public health and social work where data is often collected using self-reported surveys which are highly sensitive to e.g., the framing of the questions, the time when and place where the survey is conducted, and the level of comfort the interviewee has in sharing information with the interviewer. We propose a method for learning optimal robust classification trees based on mixed-integer robust optimization technology. In particular, we demonstrate that the problem of learning an optimal robust tree can be cast as a single-stage mixed-integer robust optimization problem with a highly nonlinear and discontinuous objective. We reformulate this problem equivalently as a two-stage linear robust optimization problem for which we devise a tailored solution procedure based on constraint generation. We evaluate the performance of our approach on numerous publicly available datasets, and compare the performance to a regularized, non-robust optimal tree. We show an increase of up to 12.48% in worst-case accuracy and of up to 4.85% in average-case accuracy across several datasets and distribution shifts from using our robust solution in comparison to the non-robust one.

Biases with respect to socially-salient attributes of individuals have been well documented in evaluation processes used in settings such as admissions and hiring. We view such an evaluation process as a transformation of a distribution of the true utility of an individual for a task to an observed distribution and model it as a solution to a loss minimization problem subject to an information constraint. Our model has two parameters that have been identified as factors leading to biases: the resource-information trade-off parameter in the information constraint and the risk-averseness parameter in the loss function. We characterize the distributions that arise from our model and study the effect of the parameters on the observed distribution. The outputs of our model enrich the class of distributions that can be used to capture variation across groups in the observed evaluations. We empirically validate our model by fitting real-world datasets and use it to study the effect of interventions in a downstream selection task. These results contribute to an understanding of the emergence of bias in evaluation processes and provide tools to guide the deployment of interventions to mitigate biases.

In many practical learning scenarios, there is a discrepancy between the training and testing distribution. For instance, when training large language models, we may have access to a training set that contains many low-quality data points from different sources and want to train a model on this dataset to perform well on a testing set that contains a few high-quality points [7, 3, 18]. An appealing way to solve this problem is data reweighting [21, 23, 26], where one attributes one weight to each data point in the training set. The weight of a training sample should reflect how much this sample resembles the testing set and helps the model perform well on it. Figure 1 illustrates the general principle. Learning the optimal weights can be cast as a bilevel optimization problem [9], where the optimal weights are such that training the model with these weights leads to the smallest test loss possible. The weights are usually constrained to sum to one, leading to an optimization problem on the simplex, which is usually solved with mirror descent [19].

The optimal robot assembly planning problem is challenging due to the necessity of finding the optimal solution amongst an exponentially vast number of possible plans, all while satisfying a selection of constraints. Traditionally, robotic assembly planning problems have been solved using heuristics, but these methods are specific to a given objective structure or set of problem parameters. In this paper, we propose a novel approach to robotic assembly planning that poses assembly sequencing as a sequential decision making problem, enabling us to harness methods that far outperform the state-of-the-art. We formulate the problem as a Markov Decision Process (MDP) and utilize Dynamic Programming (DP) to find optimal assembly policies for moderately sized strictures. We further expand our framework to exploit the deterministic nature of assembly planning and introduce a class of optimal Graph Exploration Assembly Planners (GEAPs). For larger structures, we show how Reinforcement Learning (RL) enables us to learn policies that generate high reward assembly sequences. We evaluate our approach on a variety of robotic assembly problems, such as the assembly of the Hubble Space Telescope, the International Space Station, and the James Webb Space Telescope. We further showcase how our DP, GEAP, and RL implementations are capable of finding optimal solutions under a variety of different objective functions and how our formulation allows us to translate precedence constraints to branch pruning and thus further improve performance. We have published our code at https://github.com/labicon/ORASP-Code.

Transformers are remarkably good at in-context learning (ICL) -- learning from demonstrations without parameter updates -- but how they perform ICL remains a mystery. Recent work suggests that Transformers may learn in-context by internally running Gradient Descent, a first-order optimization method. In this paper, we instead demonstrate that Transformers learn to implement higher-order optimization methods to perform ICL. Focusing on in-context linear regression, we show that Transformers learn to implement an algorithm very similar to Iterative Newton's Method, a higher-order optimization method, rather than Gradient Descent. Empirically, we show that predictions from successive Transformer layers closely match different iterations of Newton's Method linearly, with each middle layer roughly computing 3 iterations. In contrast, exponentially more Gradient Descent steps are needed to match an additional Transformers layer; this suggests that Transformers have an comparable rate of convergence with high-order methods such as Iterative Newton, which are exponentially faster than Gradient Descent. We also show that Transformers can learn in-context on ill-conditioned data, a setting where Gradient Descent struggles but Iterative Newton succeeds. Finally, we show theoretical results which support our empirical findings and have a close correspondence with them: we prove that Transformers can implement $k$ iterations of Newton's method with $\mathcal{O}(k)$ layers.

In predictive modeling with simulation or machine learning, it is critical to accurately assess the quality of estimated values through output analysis. In recent decades output analysis has become enriched with methods that quantify the impact of input data uncertainty in the model outputs to increase robustness. However, most developments are applicable assuming that the input data adheres to a parametric family of distributions. We propose a unified output analysis framework for simulation and machine learning outputs through the lens of Monte Carlo sampling. This framework provides nonparametric quantification of the variance and bias induced in the outputs with higher-order accuracy. Our new bias-corrected estimation from the model outputs leverages the extension of fast iterative bootstrap sampling and higher-order influence functions. For the scalability of the proposed estimation methods, we devise budget-optimal rules and leverage control variates for variance reduction. Our theoretical and numerical results demonstrate a clear advantage in building more robust confidence intervals from the model outputs with higher coverage probability.