Modern machine learning models are trained on diverse datasets and tasks to improve generalization. A key challenge in multitask learning is determining the optimal data mixing and sampling strategy across different data sources. Prior research in this multi-task learning setting has primarily focused on mitigating gradient conflicts between tasks. However, we observe that many real-world multitask learning scenarios--such as multilingual training and multi-domain learning in large foundation models--exhibit predominantly positive task interactions with minimal or no gradient conflict. Building on this insight, we introduce PiKE (Positive gradient interaction-based K-task weights Estimator), an adaptive data mixing algorithm that dynamically adjusts task contributions throughout training. PiKE optimizes task sampling to minimize overall loss, effectively leveraging positive gradient interactions with almost no additional computational overhead. We establish theoretical convergence guarantees for PiKE and demonstrate its superiority over static and non-adaptive mixing strategies. Additionally, we extend PiKE to promote fair learning across tasks, ensuring balanced progress and preventing task underrepresentation. Empirical evaluations on large-scale language model pretraining show that PiKE consistently outperforms existing heuristic and static mixing strategies, leading to faster convergence and improved downstream task performance.

We study procurement auctions, where an auctioneer seeks to acquire services from strategic sellers with private costs. The quality of services is measured by a submodular function known to the auctioneer. Our goal is to design computationally efficient procurement auctions that (approximately) maximize the difference between the quality of the acquired services and the total cost of the sellers, while ensuring incentive compatibility (IC), individual rationality (IR) for sellers, and non-negative surplus (NAS) for the auctioneer. Our contributions are twofold: (i) we provide an improved analysis of existing algorithms for non-positive submodular function maximization, and (ii) we design efficient frameworks that transform submodular optimization algorithms into mechanisms that are IC, IR, NAS, and approximation-preserving. These frameworks apply to both the offline setting, where all sellers' bids and services are available simultaneously, and the online setting, where sellers arrive in an adversarial order, requiring the auctioneer to make irrevocable decisions. We also explore whether state-of-the-art submodular optimization algorithms can be converted into descending auctions in adversarial settings, where the schedule of descending prices is determined by an adversary. We show that a submodular optimization algorithm satisfying bi-criteria $(1/2, 1)$-approximation in welfare can be effectively adapted to a descending auction. Additionally, we establish a connection between descending auctions and online submodular optimization. Finally, we demonstrate the practical applications of our frameworks by instantiating them with state-of-the-art submodular optimization algorithms and empirically comparing their welfare performance on publicly available datasets with thousands of sellers.

Fine-tuning language models (LMs) with the Adam optimizer often demands excessive memory, limiting accessibility. The "in-place" version of Stochastic Gradient Descent (IP-SGD) and Memory-Efficient Zeroth-order Optimizer (MeZO) have been proposed to address this. However, IP-SGD still requires substantial memory, and MeZO suffers from slow convergence and degraded final performance due to its zeroth-order nature. This paper introduces Addax, a novel method that improves both memory efficiency and performance of IP-SGD by integrating it with MeZO. Specifically, Addax computes zeroth- or first-order gradients of data points in the minibatch based on their memory consumption, combining these gradient estimates to update directions. By computing zeroth-order gradients for data points that require more memory and first-order gradients for others, Addax overcomes the slow convergence of MeZO and the excessive memory requirement of IP-SGD. Additionally, the zeroth-order gradient acts as a regularizer for the first-order gradient, further enhancing the model's final performance. Theoretically, we establish the convergence of Addax under mild assumptions, demonstrating faster convergence and less restrictive hyper-parameter choices than MeZO. Our experiments with diverse LMs and tasks show that Addax consistently outperforms MeZO regarding accuracy and convergence speed while having a comparable memory footprint. When fine-tuning OPT-13B with one A100 GPU, on average, Addax outperforms MeZO in accuracy/F1 score by 14% and runs 15x faster while using memory similar to MeZO. In our experiments on the larger OPT-30B model, on average, Addax outperforms MeZO in terms of accuracy/F1 score by >16 and runs 30x faster on a single H100 GPU. Moreover, Addax surpasses the performance of standard fine-tuning approaches, such as IP-SGD and Adam, in most tasks with significantly less memory requirement.

In digital online advertising, advertisers procure ad impressions simultaneously on multiple platforms, or so-called channels, such as Google Ads, Meta Ads Manager, etc., each of which consists of numerous ad auctions. We study how an advertiser maximizes total conversion (e.g. ad clicks) while satisfying aggregate return-on-investment (ROI) and budget constraints across all channels. In practice, an advertiser does not have control over, and thus cannot globally optimize, which individual ad auctions she participates in for each channel, and instead authorizes a channel to procure impressions on her behalf: the advertiser can only utilize two levers on each channel, namely setting a per-channel budget and per-channel target ROI. In this work, we first analyze the effectiveness of each of these levers for solving the advertiser's global multi-channel problem. We show that when an advertiser only optimizes over per-channel ROIs, her total conversion can be arbitrarily worse than what she could have obtained in the global problem. Further, we show that the advertiser can achieve the global optimal conversion when she only optimizes over per-channel budgets. In light of this finding, under a bandit feedback setting that mimics real-world scenarios where advertisers have limited information on ad auctions in each channels and how channels procure ads, we present an efficient learning algorithm that produces per-channel budgets whose resulting conversion approximates that of the global optimal problem. Finally, we argue that all our results hold for both single-item and multi-item auctions from which channels procure impressions on advertisers' behalf.

It is well-known that the Gale-Shapley algorithm is not truthful for all agents. Previous studies in this category concentrate on manipulations using incomplete preference lists by a single woman and by the set of all women. Little is known about manipulations by a subset of women. In this paper, we consider manipulations by any subset of women with arbitrary preferences. We show that a strong Nash equilibrium of the induced manipulation game always exists among the manipulators and the equilibrium outcome is unique and Pareto-dominant. In addition, the set of matchings achievable by manipulations has a lattice structure. We also examine the super-strong Nash equilibrium in the end.

A paper by Deng and Conitzer in AAAI'17 introduces disarmament games, in which players alternatingly commit not to play certain pure strategies. However, in practice, disarmament usually does not consist in removing a strategy, but rather in removing a resource (and doing so rules out all the strategies in which that resource is used simultaneously). In this paper, we introduce a model of disarmament games in which resources, rather than strategies, are removed. We prove NP-completeness of several formulations of the problem of achieving desirable outcomes via disarmament. We then study the case where resources can be fractionally removed, and prove a result analogous to the folk theorem that all desirable outcomes can be achieved. We show that we can approximately achieve any desirable outcome in a polynomial number of rounds, though determining whether a given outcome can be obtained in a given number of rounds remains NP-complete.

Developing efficient and guaranteed nonconvex algorithms has been an important challenge in modern machine learning. Algorithms with good empirical performance such as stochastic gradient descent often lack theoretical guarantees. In this paper, we analyze the class of homotopy or continuation methods for global optimization of nonconvex functions. These methods start from an objective function that is efficient to optimize (e.g. convex), and progressively modify it to obtain the required objective, and the solutions are passed along the homotopy path. For the challenging problem of tensor PCA, we prove global convergence of the homotopy method in the "high noise" regime. The signal-to-noise requirement for our algorithm is tight in the sense that it matches the recovery guarantee for the best degree-4 sum-of-squares algorithm. In addition, we prove a phase transition along the homotopy path for tensor PCA. This allows to simplify the homotopy method to a local search algorithm, viz., tensor power iterations, with a specific initialization and a noise injection procedure, while retaining the theoretical guarantees.

The generality of decision and game theory has enabled domain-independent progress in AI research. For example, a better algorithm for finding good policies in (PO)MDPs can be instantly used in a variety of applications. But such a general theory is lacking when it comes to moral decision making. For AI applications with a moral component, are we then forced to build systems based on many ad-hoc rules? In this paper we discuss possible ways to avoid this conclusion.

In various matching market settings, such as hospital-doctor matching markets (Hatfield and Milgrom 2005), the existence of stable outcomes depends on substitutability of preferences. But can these stable matchings be computed efficiently, as in the one-to-one matching case? The algorithm of (Hatfield and Milgrom 2005) requires efficient implementation of a choice function over substitutable preferences. We show that even given efficient access to a value oracle or preference relation satisfying substitutability, exponentially many queries may be required in the worst case to implement a choice function. Indeed, this extends to examples where a stable matching requires exponential time to compute. We characterize the computational complexity of stable matchings by showing that efficient computation of a choice function is equivalent to efficient verification—determining whether or not, for a given set, the most preferred subset is the entire set itself. Clearly, verification is necessary for computation, but we show that it is also sufficient: specifically, given a verifier, we design a polynomial-time algorithm for computing a choice function, implying an efficient algorithm for stable matching. We then show that a verifier can be implemented efficiently for various classes of functions, such as submodular functions, implying efficient stable matching algorithms for a broad range of settings. We also investigate the effect of ties in the preference order, which causes complications both in defining substitutes and in computation. In this case, we tightly connect the computational complexity of the choice function to a measure on the number of ties.

Much recent work in the AI community concerns algorithms for computing optimal mixed strategies to commit to, as well as the deployment of such algorithms in real security applications. Another possibility is to commit not to play certain actions. If only one player makes such a commitment, then this is generally less powerful than completely committing to a single mixed strategy. However, if players can alternatingly commit not to play certain actions and thereby iteratively reduce their strategy spaces, then desirable outcomes can be obtained that would not have been possible with just a single player committing to a mixed strategy. We refer to such a setting as a disarmament game. In this paper, we study disarmament for two-player normal-form games. We show that deciding whether an outcome can be obtained with disarmament is NP-complete (even for a fixed number of rounds), if only pure strategies can be removed. On the other hand, for the case where mixed strategies can be removed, we provide a folk theorem that shows that all desirable utility profiles can be obtained, and give an efficient algorithm for (approximately) obtaining them.