arXiv subjects

Nika Haghtalab

Publications and source records attributed to Nika Haghtalab.

At least 37 records · Page 2Linked to original sources

arXiv2025

Panprediction: Optimal Predictions for Any Downstream Task and Loss

Supervised learning is classically formulated as training a model to minimize a fixed loss function over a fixed distribution, or task. However, an emerging paradigm instead views model training as extracting enough information from data so that the model can be used to minimize many losses on many downstream tasks. We formalize a mathematical framework for this paradigm, which we call panprediction, and study its statistical complexity. Formally, panprediction generalizes omniprediction and sits upstream from multi-group learning, which respectively focus on predictions that generalize to many downstream losses or many downstream tasks, but not both. Concretely, we design algorithms that learn deterministic and randomized panpredictors with $\tilde{O}(1/\varepsilon^3)$ and $\tilde{O}(1/\varepsilon^2)$ samples, respectively. Our results demonstrate that under mild assumptions, simultaneously minimizing infinitely many losses on infinitely many tasks can be as statistically easy as minimizing one loss on one task. Along the way, we improve the best known sample complexity guarantee of deterministic omniprediction by a factor of $1/\varepsilon$, and match all other known sample complexity guarantees of omniprediction and multi-group learning. Our key technical ingredient is a nearly lossless reduction from panprediction to a statistically efficient notion of calibration, called step calibration.

cs.LG↗

arXiv2025

Distortion of AI Alignment: Does Preference Optimization Optimize for Preferences?

After pre-training, large language models are aligned with human preferences based on pairwise comparisons. State-of-the-art alignment methods (such as PPO-based RLHF and DPO) are built on the assumption of aligning with a single preference model, despite being deployed in settings where users have diverse preferences. As a result, it is not even clear that these alignment methods produce models that satisfy users on average -- a minimal requirement for pluralistic alignment. Drawing on social choice theory and modeling users' comparisons through individual Bradley-Terry (BT) models, we introduce an alignment method's distortion: the worst-case ratio between the optimal achievable average utility, and the average utility of the learned policy. The notion of distortion helps draw sharp distinctions between alignment methods: Nash Learning from Human Feedback achieves the minimax optimal distortion of $(\frac{1}{2} + o(1)) \cdot β$ (for the BT temperature $β$), robustly across utility distributions, distributions of comparison pairs, and permissible KL divergences from the reference policy. RLHF and DPO, by contrast, suffer $\geq (1 - o(1)) \cdot β$ distortion already without a KL constraint, and $e^{Ω(β)}$ or even unbounded distortion in the full setting, depending on how comparison pairs are sampled.

cs.LG↗

arXiv2025

Learning in Stackelberg Games with Non-myopic Agents

We study Stackelberg games where a principal repeatedly interacts with a non-myopic long-lived agent, without knowing the agent's payoff function. Although learning in Stackelberg games is well-understood when the agent is myopic, dealing with non-myopic agents poses additional complications. In particular, non-myopic agents may strategize and select actions that are inferior in the present in order to mislead the principal's learning algorithm and obtain better outcomes in the future. We provide a general framework that reduces learning in presence of non-myopic agents to robust bandit optimization in the presence of myopic agents. Through the design and analysis of minimally reactive bandit algorithms, our reduction trades off the statistical efficiency of the principal's learning algorithm against its effectiveness in inducing near-best-responses. We apply this framework to Stackelberg security games (SSGs), pricing with unknown demand curve, general finite Stackelberg games, and strategic classification. In each setting, we characterize the type and impact of misspecifications present in near-best responses and develop a learning algorithm robust to such misspecifications. On the way, we improve the state-of-the-art query complexity of learning in SSGs with $n$ targets from $O(n^3)$ to a near-optimal $\widetilde{O}(n)$ by uncovering a fundamental structural property of these games. The latter result is of independent interest beyond learning with non-myopic agents.

cs.GT↗

arXiv2025

From Style to Facts: Mapping the Boundaries of Knowledge Injection with Finetuning

Finetuning provides a scalable and cost-effective means of customizing language models for specific tasks or response styles, with greater reliability than prompting or in-context learning. In contrast, the conventional wisdom is that injecting knowledge via finetuning results in brittle performance and poor generalization. We argue that the dichotomy of "task customization" (e.g., instruction tuning) and "knowledge injection" (e.g., teaching new facts) is a distinction without a difference. We instead identify concrete factors that explain the heterogeneous effectiveness observed with finetuning. To this end, we conduct a large-scale experimental study of finetuning the frontier Gemini v1.5 model family on a spectrum of datasets that are artificially engineered to interpolate between the strengths and failure modes of finetuning. Our findings indicate that question-answer training data formats provide much stronger knowledge generalization than document/article-style training data, numerical information can be harder for finetuning to retain than categorical information, and models struggle to apply finetuned knowledge during multi-step reasoning even when trained on similar examples -- all factors that render "knowledge injection" to be especially difficult, even after controlling for considerations like data augmentation and information volume. On the other hand, our findings also indicate that it is not fundamentally more difficult to finetune information about a real-world event than information about what a model's writing style should be.

cs.CL↗

arXiv2025

Algorithmic Content Selection and the Impact of User Disengagement

Digital services face a fundamental trade-off in content selection: they must balance the immediate revenue gained from high-reward content against the long-term benefits of maintaining user engagement. Traditional multi-armed bandit models assume that users remain perpetually engaged, failing to capture the possibility that users may disengage when dissatisfied, thereby reducing future revenue potential. In this work, we introduce a model for the content selection problem that explicitly accounts for variable user engagement and disengagement. In our framework, content that maximizes immediate reward is not necessarily optimal in terms of fostering sustained user engagement. Our contributions are twofold. First, we develop computational and statistical methods for offline optimization and online learning of content selection policies. For users whose engagement patterns are defined by $k$ distinct levels, we design a dynamic programming algorithm that computes the exact optimal policy in $O(k^2)$ time. Moreover, we derive no-regret learning guarantees for an online learning setting in which the platform serves a series of users with unknown and potentially adversarial engagement patterns. Second, we introduce the concept of modified demand elasticity which captures how small changes in a user's overall satisfaction affect the platform's ability to secure long-term revenue. This notion generalizes classical demand elasticity by incorporating the dynamics of user re-engagement, thereby revealing key insights into the interplay between engagement and revenue. Notably, our analysis uncovers a counterintuitive phenomenon: although higher friction (i.e., a reduced likelihood of re-engagement) typically lowers overall revenue, it can simultaneously lead to higher user engagement under optimal content selection policies.

cs.LG↗

arXiv2025

Multi-Agent Risks from Advanced AI

The rapid development of advanced AI agents and the imminent deployment of many instances of these agents will give rise to multi-agent systems of unprecedented complexity. These systems pose novel and under-explored risks. In this report, we provide a structured taxonomy of these risks by identifying three key failure modes (miscoordination, conflict, and collusion) based on agents' incentives, as well as seven key risk factors (information asymmetries, network effects, selection pressures, destabilising dynamics, commitment problems, emergent agency, and multi-agent security) that can underpin them. We highlight several important instances of each risk, as well as promising directions to help mitigate them. By anchoring our analysis in a range of real-world examples and experimental evidence, we illustrate the distinct challenges posed by multi-agent systems and their implications for the safety, governance, and ethics of advanced AI.

cs.MA↗

arXiv2024

Learning With Multi-Group Guarantees For Clusterable Subpopulations

A canonical desideratum for prediction problems is that performance guarantees should hold not just on average over the population, but also for meaningful subpopulations within the overall population. But what constitutes a meaningful subpopulation? In this work, we take the perspective that relevant subpopulations should be defined with respect to the clusters that naturally emerge from the distribution of individuals for which predictions are being made. In this view, a population refers to a mixture model whose components constitute the relevant subpopulations. We suggest two formalisms for capturing per-subgroup guarantees: first, by attributing each individual to the component from which they were most likely drawn, given their features; and second, by attributing each individual to all components in proportion to their relative likelihood of having been drawn from each component. Using online calibration as a case study, we study a multi-objective algorithm that provides guarantees for each of these formalisms by handling all plausible underlying subpopulation structures simultaneously, and achieve an $O(T^{1/2})$ rate even when the subpopulations are not well-separated. In comparison, the more natural cluster-then-predict approach that first recovers the structure of the subpopulations and then makes predictions suffers from a $O(T^{2/3})$ rate and requires the subpopulations to be separable. Along the way, we prove that providing per-subgroup calibration guarantees for underlying clusters can be easier than learning the clusters: separation between median subgroup features is required for the latter but not the former.

cs.LG↗

arXiv2024

Delegating Data Collection in Decentralized Machine Learning

Motivated by the emergence of decentralized machine learning (ML) ecosystems, we study the delegation of data collection. Taking the field of contract theory as our starting point, we design optimal and near-optimal contracts that deal with two fundamental information asymmetries that arise in decentralized ML: uncertainty in the assessment of model quality and uncertainty regarding the optimal performance of any model. We show that a principal can cope with such asymmetry via simple linear contracts that achieve 1-1/e fraction of the optimal utility. To address the lack of a priori knowledge regarding the optimal performance, we give a convex program that can adaptively and efficiently compute the optimal contract. We also study linear contracts and derive the optimal utility in the more complex setting of multiple interactions.

cs.LG↗

arXiv2024

Platforms for Efficient and Incentive-Aware Collaboration

Collaboration is crucial for reaching collective goals. However, its effectiveness is often undermined by the strategic behavior of individual agents -- a fact that is captured by a high Price of Stability (PoS) in recent literature [Blum et al., 2021]. Implicit in the traditional PoS analysis is the assumption that agents have full knowledge of how their tasks relate to one another. We offer a new perspective on bringing about efficient collaboration among strategic agents using information design. Inspired by the growing importance of collaboration in machine learning (such as platforms for collaborative federated learning and data cooperatives), we propose a framework where the platform has more information about how the agents' tasks relate to each other than the agents themselves. We characterize how and to what degree such platforms can leverage their information advantage to steer strategic agents toward efficient collaboration. Concretely, we consider collaboration networks where each node is a task type held by one agent, and each task benefits from contributions made in their inclusive neighborhood of tasks. This network structure is known to the agents and the platform, but only the platform knows each agent's real location -- from the agents' perspective, their location is determined by a random permutation. We employ private Bayesian persuasion and design two families of persuasive signaling schemes that the platform can use to ensure a small total workload when agents follow the signal. The first family aims to achieve the minmax optimal approximation ratio compared to the optimal collaboration, which is shown to be $Θ(\sqrt{n})$ for unit-weight graphs, $Θ(n^{2/3})$ for graphs with constant minimum edge weights, and $O(n^{3/4})$ for general weighted graphs. The second family ensures per-instance strict improvement compared to full information disclosure.

cs.GT↗

arXiv2024

Truthfulness of Calibration Measures

We initiate the study of the truthfulness of calibration measures in sequential prediction. A calibration measure is said to be truthful if the forecaster (approximately) minimizes the expected penalty by predicting the conditional expectation of the next outcome, given the prior distribution of outcomes. Truthfulness is an important property of calibration measures, ensuring that the forecaster is not incentivized to exploit the system with deliberate poor forecasts. This makes it an essential desideratum for calibration measures, alongside typical requirements, such as soundness and completeness. We conduct a taxonomy of existing calibration measures and their truthfulness. Perhaps surprisingly, we find that all of them are far from being truthful. That is, under existing calibration measures, there are simple distributions on which a polylogarithmic (or even zero) penalty is achievable, while truthful prediction leads to a polynomial penalty. Our main contribution is the introduction of a new calibration measure termed the Subsampled Smooth Calibration Error (SSCE) under which truthful prediction is optimal up to a constant multiplicative factor.

cs.LG↗

arXiv2024

Is Knowledge Power? On the (Im)possibility of Learning from Strategic Interactions

When learning in strategic environments, a key question is whether agents can overcome uncertainty about their preferences to achieve outcomes they could have achieved absent any uncertainty. Can they do this solely through interactions with each other? We focus this question on the ability of agents to attain the value of their Stackelberg optimal strategy and study the impact of information asymmetry. We study repeated interactions in fully strategic environments where players' actions are decided based on learning algorithms that take into account their observed histories and knowledge of the game. We study the pure Nash equilibria (PNE) of a meta-game where players choose these algorithms as their actions. We demonstrate that if one player has perfect knowledge about the game, then any initial informational gap persists. That is, while there is always a PNE in which the informed agent achieves her Stackelberg value, there is a game where no PNE of the meta-game allows the partially informed player to achieve her Stackelberg value. On the other hand, if both players start with some uncertainty about the game, the quality of information alone does not determine which agent can achieve her Stackelberg value. In this case, the concept of information asymmetry becomes nuanced and depends on the game's structure. Overall, our findings suggest that repeated strategic interactions alone cannot facilitate learning effectively enough to earn an uninformed player her Stackelberg value.

cs.GT↗

arXiv2024

Smooth Nash Equilibria: Algorithms and Complexity

A fundamental shortcoming of the concept of Nash equilibrium is its computational intractability: approximating Nash equilibria in normal-form games is PPAD-hard. In this paper, inspired by the ideas of smoothed analysis, we introduce a relaxed variant of Nash equilibrium called $σ$-smooth Nash equilibrium, for a smoothness parameter $σ$. In a $σ$-smooth Nash equilibrium, players only need to achieve utility at least as high as their best deviation to a $σ$-smooth strategy, which is a distribution that does not put too much mass (as parametrized by $σ$) on any fixed action. We distinguish two variants of $σ$-smooth Nash equilibria: strong $σ$-smooth Nash equilibria, in which players are required to play $σ$-smooth strategies under equilibrium play, and weak $σ$-smooth Nash equilibria, where there is no such requirement. We show that both weak and strong $σ$-smooth Nash equilibria have superior computational properties to Nash equilibria: when $σ$ as well as an approximation parameter $ε$ and the number of players are all constants, there is a constant-time randomized algorithm to find a weak $ε$-approximate $σ$-smooth Nash equilibrium in normal-form games. In the same parameter regime, there is a polynomial-time deterministic algorithm to find a strong $ε$-approximate $σ$-smooth Nash equilibrium in a normal-form game. These results stand in contrast to the optimal algorithm for computing $ε$-approximate Nash equilibria, which cannot run in faster than quasipolynomial-time. We complement our upper bounds by showing that when either $σ$ or $ε$ is an inverse polynomial, finding a weak $ε$-approximate $σ$-smooth Nash equilibria becomes computationally intractable.

cs.GT↗

arXiv2024

Communicating with Anecdotes

We study a communication game between a sender and a receiver. The sender chooses one of her signals about the state of the world (i.e., anecdotes) and communicates to the receiver who takes an action affecting both players. The sender and the receiver both care about the state of the world but are also influenced by personal preferences, so their ideal actions can differ. We characterize perfect Bayesian equilibria. The sender faces a temptation to persuade: she wants to select a biased anecdote to influence the receiver's action. Anecdotes are still informative to the receiver (who will debias at equilibrium) but the attempt to persuade comes at a cost to precision. This gives rise to informational homophily where the receiver prefers to listen to like-minded senders because they provide higher-precision signals. Communication becomes polarized when the sender is an expert with access to many signals, with the sender choosing extreme outlier anecdotes at equilibrium (unless preferences are perfectly aligned). This polarization dissipates all gains from communication with an increasingly well-informed sender when the anecdote distribution is heavy-tailed. Experts can therefore face a curse of informedness: receivers will prefer to listen to less-informed senders who cannot pick biased signals as easily.

econ.TH↗

arXiv2024

Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation

Black-box finetuning is an emerging interface for adapting state-of-the-art language models to user needs. However, such access may also let malicious actors undermine model safety. To demonstrate the challenge of defending finetuning interfaces, we introduce covert malicious finetuning, a method to compromise model safety via finetuning while evading detection. Our method constructs a malicious dataset where every individual datapoint appears innocuous, but finetuning on the dataset teaches the model to respond to encoded harmful requests with encoded harmful responses. Applied to GPT-4, our method produces a finetuned model that acts on harmful instructions 99% of the time and avoids detection by defense mechanisms such as dataset inspection, safety evaluations, and input/output classifiers. Our findings question whether black-box finetuning access can be secured against sophisticated adversaries.

cs.CR↗

arXiv2024

On-Demand Sampling: Learning Optimally from Multiple Distributions

Social and real-world considerations such as robustness, fairness, social welfare and multi-agent tradeoffs have given rise to multi-distribution learning paradigms, such as collaborative learning, group distributionally robust optimization, and fair federated learning. In each of these settings, a learner seeks to uniformly minimize its expected loss over $n$ predefined data distributions, while using as few samples as possible. In this paper, we establish the optimal sample complexity of these learning paradigms and give algorithms that meet this sample complexity. Importantly, our sample complexity bounds for multi-distribution learning exceed that of learning a single distribution by only an additive factor of $n \log(n) / ε^2$. This improves upon the best known sample complexity bounds for fair federated learning by Mohri et al. and collaborative learning by Nguyen and Zakynthinou by multiplicative factors of $n$ and $\log(n)/ε^3$, respectively. We also provide the first sample complexity bounds for the group DRO objective of Sagawa et al. To guarantee these optimal sample complexity bounds, our algorithms learn to sample from data distributions on demand. Our algorithm design and analysis are enabled by our extensions of online learning techniques for solving stochastic zero-sum games. In particular, we contribute stochastic variants of no-regret dynamics that can trade off between players' differing sampling costs.

cs.LG↗

arXiv2024

Improved Bayes Risk Can Yield Reduced Social Welfare Under Competition

As the scale of machine learning models increases, trends such as scaling laws anticipate consistent downstream improvements in predictive accuracy. However, these trends take the perspective of a single model-provider in isolation, while in reality providers often compete with each other for users. In this work, we demonstrate that competition can fundamentally alter the behavior of these scaling trends, even causing overall predictive accuracy across users to be non-monotonic or decreasing with scale. We define a model of competition for classification tasks, and use data representations as a lens for studying the impact of increases in scale. We find many settings where improving data representation quality (as measured by Bayes risk) decreases the overall predictive accuracy across users (i.e., social welfare) for a marketplace of competing model-providers. Our examples range from closed-form formulas in simple settings to simulations with pretrained representations on CIFAR-10. At a conceptual level, our work suggests that favorable scaling trends for individual model-providers need not translate to downstream improvements in social welfare in marketplaces with multiple model providers.

cs.GT↗

arXiv2024

Can Probabilistic Feedback Drive User Impacts in Online Platforms?

A common explanation for negative user impacts of content recommender systems is misalignment between the platform's objective and user welfare. In this work, we show that misalignment in the platform's objective is not the only potential cause of unintended impacts on users: even when the platform's objective is fully aligned with user welfare, the platform's learning algorithm can induce negative downstream impacts on users. The source of these user impacts is that different pieces of content may generate observable user reactions (feedback information) at different rates; these feedback rates may correlate with content properties, such as controversiality or demographic similarity of the creator, that affect the user experience. Since differences in feedback rates can impact how often the learning algorithm engages with different content, the learning algorithm may inadvertently promote content with certain such properties. Using the multi-armed bandit framework with probabilistic feedback, we examine the relationship between feedback rates and a learning algorithm's engagement with individual arms for different no-regret algorithms. We prove that no-regret algorithms can exhibit a wide range of dependencies: if the feedback rate of an arm increases, some no-regret algorithms engage with the arm more, some no-regret algorithms engage with the arm less, and other no-regret algorithms engage with the arm approximately the same number of times. From a platform design perspective, our results highlight the importance of looking beyond regret when measuring an algorithm's performance, and assessing the nature of a learning algorithm's engagement with different types of content as well as their resulting downstream impacts.

cs.LG↗

arXiv2023

A Unifying Perspective on Multi-Calibration: Game Dynamics for Multi-Objective Learning

We provide a unifying framework for the design and analysis of multicalibrated predictors. By placing the multicalibration problem in the general setting of multi-objective learning -- where learning guarantees must hold simultaneously over a set of distributions and loss functions -- we exploit connections to game dynamics to achieve state-of-the-art guarantees for a diverse set of multicalibration learning problems. In addition to shedding light on existing multicalibration guarantees and greatly simplifying their analysis, our approach also yields improved guarantees, such as obtaining stronger multicalibration conditions that scale with the square-root of group size and improving the complexity of $k$-class multicalibration by an exponential factor of $k$. Beyond multicalibration, we use these game dynamics to address emerging considerations in the study of group fairness and multi-distribution learning.

cs.LG↗