High dialogue engagement is a crucial indicator of an effective conversation. A reliable measure of engagement could help benchmark large language models, enhance the effectiveness of human-computer interactions, or improve personal communication skills. However, quantifying engagement is challenging, since it is subjective and lacks a ''gold standard''. This paper proposes PMIScore, an efficient unsupervised approach to quantify dialogue engagement. It uses pointwise mutual information (PMI), which is the probability of generating a response conditioning on the conversation history. Thus, PMIScore offers a clear interpretation of engagement. As directly computing PMI is intractable due to the complexity of dialogues, PMIScore learned it through a dual form of divergence. The algorithm includes generating positive and negative dialogue pairs, extracting embeddings by large language models (LLMs), and training a small neural network using a mutual information loss function. We validated PMIScore on both synthetic and real-world datasets. Our results demonstrate the effectiveness of PMIScore in PMI estimation and the reasonableness of the PMI metric itself.

We introduce the GEM (Generative Estimator for Mutual Information), an evaluation metric for assessing language generation by large language models (LLMs), particularly in generating informative judgments, without the need for a gold standard reference. GEM broadens the scenarios where we can benchmark LLM generation performance-from traditional ones, like machine translation and summarization, where gold standard references are readily available, to subjective tasks without clear gold standards, such as academic peer review. GEM uses a generative model to estimate mutual information between candidate and reference responses, without requiring the reference to be a gold standard. In experiments on two human-annotated datasets, GEM demonstrates competitive correlations with human scores compared to the state-of-the-art GPT-4o Examiner, and outperforms all other baselines. Additionally, GEM is more robust against strategic manipulation, such as rephrasing or elongation, which can artificially inflate scores under a GPT-4o Examiner. We also present GRE-bench (Generating Review Evaluation Benchmark) which evaluates LLMs based on how well they can generate high-quality peer reviews for academic research papers. Because GRE-bench is based upon GEM, it inherits its robustness properties. Additionally, GRE-bench circumvents data contamination problems (or data leakage) by using the continuous influx of new open-access research papers and peer reviews each year. We show GRE-bench results of various popular LLMs on their peer review capabilities using the ICLR2023 dataset.

Forecast aggregation combines the predictions of multiple forecasters to improve accuracy. However, the lack of knowledge about forecasters' information structure hinders optimal aggregation. Given a family of information structures, robust forecast aggregation aims to find the aggregator with minimal worst-case regret compared to the omniscient aggregator. Previous approaches for robust forecast aggregation rely on heuristic observations and parameter tuning. We propose an algorithmic framework for robust forecast aggregation. Our framework provides efficient approximation schemes for general information aggregation with a finite family of possible information structures. In the setting considered by Arieli et al. [2018] where two agents receive independent signals conditioned on a binary state, our framework also provides efficient approximation schemes by imposing Lipschitz conditions on the aggregator or discrete conditions on agents' reports. Numerical experiments demonstrate the effectiveness of our method by providing a nearly optimal aggregator in the setting considered by Arieli et al. [2018].

The standard framework of online bidding algorithm design assumes that the seller commits himself to faithfully implementing the rules of the adopted auction.However, the seller may attempt to cheat in execution to increase his revenue if the auction belongs to the class of non-credible auctions.For example, in a second-price auction, the seller could create a fake bid between the highest bid and the second highest bid.This paper focuses on one such case of online bidding in repeated second-price auctions.At each time 𝑡, the winner with bid 𝑏 𝑡 is charged not the highest competing bid 𝑑 𝑡 but a manipulated price 𝑝 𝑡 = 𝛼 0 𝑑 𝑡 + (1 -𝛼 0 )𝑏 𝑡 , where the parameter 𝛼 0 ∈ [0, 1] in essence measures the seller's credibility.Unlike classic repeated-auction settings where the bidder has access to samples (𝑑 𝑠 ) 𝑡 -1 𝑠=1 , she can only receive mixed signals of𝑠=1 and 𝛼 0 in this problem.The task for the bidder is to learn not only the bid distributions of her competitors but also the seller's credibility.We establish regret lower bounds in various information models and provide corresponding online bidding algorithms that can achieve near-optimal performance.Specifically, * Both authors contributed equally to this research.

We consider a robust aggregation problem in the presence of both truthful and adversarial experts. The truthful experts will report their private signals truthfully, while the adversarial experts can report arbitrarily. We assume experts are marginally symmetric in the sense that they share the same common prior and marginal posteriors. The rule maker needs to design an aggregator to predict the true world state from these experts' reports, without knowledge of the underlying information structures or adversarial strategies. We aim to find the optimal aggregator that outputs a forecast minimizing regret under the worst information structure and adversarial strategies. The regret is defined by the difference in expected loss between the aggregator and a benchmark who aggregates optimally given the information structure and reports of truthful experts.

Rating aggregation plays a crucial role in various fields, such as product recommendations, hotel rankings, and teaching evaluations. However, traditional averaging methods can be affected by participation bias, where some raters do not participate in the rating process, leading to potential distortions. In this paper, we consider a robust rating aggregation task under the participation bias. We assume that raters may not reveal their ratings with a certain probability depending on their individual ratings, resulting in partially observed samples. Our goal is to minimize the expected squared loss between the aggregated ratings and the average of all underlying ratings (possibly unobserved) in the worst-case scenario.

Peer prediction mechanisms motivate high-quality feedback with provable guarantees. However, current methods only apply to rather simple reports, like multiple-choice or scalar numbers. We aim to broaden these techniques to the larger domain of text-based reports, drawing on the recent developments in large language models (LLMs). This vastly increases the applicability of peer prediction mechanisms as textual feedback is the norm in a large variety of feedback channels: peer reviews, e-commerce customer reviews, and comments on social media.

Robust aggregation integrates predictions from multiple experts without knowledge of the experts' information structures. Prior work assumes experts are Bayesian, providing predictions as perfect posteriors based on their signals. However, real-world experts often deviate systematically from Bayesian reasoning. Our work considers experts who tend to ignore the base rate. We find that a certain degree of base rate neglect helps with robust forecast aggregation.

We consider a decision aggregation problem with two experts who each make a binary recommendation after observing a private signal about an unknown binary world state. An agent, who does not know the joint information structure between signals and states, sees the experts' recommendations and aims to match the action with the true state. Under the scenario, we study whether supplemented additionally with second-order information (each expert's forecast on the other's recommendation) could enable a better aggregation.

1 In the setting where participants are asked multiple similar possibly subjective multi-choice questions (e.g., Do you like Panda Express? Y/N; Do you like Chick-fil-A? Y/N), a series of peer prediction mechanisms have been designed to incentivize honest reports and some of them achieve dominantly truthfulness: Truth-telling is a dominant strategy and strictly dominates other “non-permutation strategy” with some mild conditions. However, those mechanisms require the participants to perform an infinite number of tasks. When the participants perform a finite number of tasks, these mechanisms only achieve approximated dominant truthfulness. The existence of a dominantly truthful multi-task peer prediction mechanism that only requires a finite number of tasks remains to be an open question that may have a negative result, even with full prior knowledge. This article answers this open question by proposing a family of mechanisms, VMI-Mechanisms, that are dominantly truthful with a finite number of tasks. A special case of this family, DMI-Mechanism, only requires ≥ 2 C tasks where C is the number of choices for each question ( C =2 for binary-choice questions). The implementation of these mechanisms does not require any prior knowledge (detail-free) and only requires ≥ 2 participants. To the best of our knowledge, any mechanism of the family is the first dominantly truthful peer prediction mechanism that works for a finite number of tasks. The core of these new mechanisms is a new family of information-monotone information measures: volume mutual information (VMI). VMI is based on a simple geometric information measure design method, the volume method. The volume method measures the informativeness of an object by “counting” the number of objects that are less informative than it. In other words, the more objects that the object of interest dominates, the more informative it is considered to be. Finally, in the setting where agents need to invest efforts to obtain their private signals, we show how to select the mechanism to optimally incentivize efforts among a proper set of VMI-Mechanisms.

Budget management strategies in repeated auctions have received growing attention in online advertising markets. However, previous work on budget management in online bidding mainly focused on second-price auctions. The rapid shift from second-price auctions to first-price auctions for online ads in recent years has motivated the challenging question of how to bid in repeated first-price auctions while controlling budgets. In this work, we study the problem of learning in repeated first-price auctions with budgets. We design a dual-based algorithm that can achieve a near-optimal $\widetilde{O}(\sqrt{T})$ regret with full information feedback where the maximum competing bid is always revealed after each auction. We further consider the setting with one-sided information feedback where only the winning bid is revealed after each auction. We show that our modified algorithm can still achieve an $\widetilde{O}(\sqrt{T})$ regret with mild assumptions on the bidder's value distribution. Finally, we complement the theoretical results with numerical experiments to confirm the effectiveness of our budget management policy.

We study a setting where Bayesian agents with a common prior have private information related to an event’s outcome and sequentially make public announcements relating to their information. Our main result shows that when agents' private information is independent conditioning on the event’s outcome whenever agents have similar beliefs about the outcome, their information is aggregated. That is, there is no false consensus. Our main result has a short proof based on a natural information-theoretic framework. A key ingredient of the framework is the equivalence between the sign of the "interaction information" and a super/sub-additive property of the value of people’s information. This provides an intuitive interpretation and an interesting application of the interaction information, which measures the amount of information shared by three random variables. We illustrate the power of this information-theoretic framework by reproving two additional results within it: 1) that agents quickly agree when announcing (summaries of) beliefs in round-robin fashion [Aaronson 2005], and 2) results from [Chen et al 2010] on when prediction market agents should release information to maximize their payment. We also interpret the information-theoretic framework and the above results in prediction markets by proving that the expected reward of revealing information is the conditional mutual information of the information revealed.

Peer review lies at the core of the academic process, but even well-intentioned reviewers can still provide noisy ratings. While ranking papers by average ratings may reduce noise, varying noise levels and systematic biases stemming from ``cheap'' signals (e.g. author identity, proof length) can lead to unfairness. Detecting and correcting bias is challenging, as ratings are subjective and unverifiable. Unlike previous works relying on prior knowledge or historical data, we propose a one-shot noise calibration process without any prior information. We ask reviewers to predict others' scores and use these predictions for calibration. Assuming reviewers adjust their predictions according to the noise, we demonstrate that the calibrated score results in a more robust ranking compared to average ratings, even with varying noise levels and biases. In detail, we show that the error probability of the calibrated score approaches zero as the number of reviewers increases and is significantly lower compared to average ratings when the number of reviewers is small.

A/B testing, or controlled experiments, is the gold standard approach to causally compare the performance of algorithms on online platforms. However, conventional Bernoulli randomization in A/B testing faces many challenges such as spillover and carryover effects. Our study focuses on another challenge, especially for A/B testing on two-sided platforms – budget constraints. Buyers on two-sided platforms often have limited budgets, where the conventional A/B testing may be infeasible to be applied, partly because two variants of allocation algorithms may conflict and lead some buyers to exceed their budgets if they are implemented simultaneously. We develop a model to describe two-sided platforms where buyers have limited budgets. We then provide an optimal experimental design that guarantees small bias and minimum variance. Bias is lower when there is more budget and a higher supply-demand rate. We test our experimental design on both synthetic data and real-world data, which verifies the theoretical results and shows our advantage compared to Bernoulli randomization.

In the setting where we ask participants multiple similar possibly subjective multi-choice questions (e.g. Do you like Bulbasaur? Y/N; do you like Squirtle? Y/N), peer prediction aims to design mechanisms that encourage honest feedback without verification. A series of works have successfully designed multi-task peer prediction mechanisms where reporting truthfully is better than any other strategy (dominantly truthful), while they require an infinite number of tasks. A recent work proposes the first multi-task peer prediction mechanism, Determinant Mutual Information (DMI)-Mechanism, where not only is dominantly truthful but also works for a finite number of tasks (practical). However, the existence of other practical dominantly-truthful multi-task peer prediction mechanisms remains to be an open question. This work answers the above question by providing - a new family of information-monotone information measures: volume mutual information (VMI), where DMI is a special case; - a new family of practical dominantly-truthful multi-task peer prediction mechanisms, VMI-Mechanisms. To illustrate the importance of VMI-Mechanisms, we also provide a tractable effort incentive optimization goal. We show that DMI-Mechanism may not be not optimal but we can construct a sequence of VMI-Mechanisms that are approximately optimal. The main technical highlight in this paper is a novel geometric information measure, Volume Mutual Information, that is based on a simple idea: we can measure an object A’s information amount by the number of objects that is less informative than A. Different densities over the object lead to different information measures. This also gives Determinant Mutual Information a simple geometric interpretation.

When we use the wisdom of the crowds, we usually rank the answers according to their popularity, especially when we cannot verify the answers. However, this can be very dangerous when the majority make systematic mistakes. A fundamental question arises: can we build a hierarchy among the answers without any prior where the higher-ranking answers, which may not be supported by the majority, are from more sophisticated people? To address the question, we propose 1) a novel model to describe people's thinking hierarchy; 2) two algorithms to learn the thinking hierarchy without any prior; 3) a novel open-response based crowdsourcing approach based on the above theoretic framework. In addition to theoretic justifications, we conduct four empirical crowdsourcing studies and show that a) the accuracy of the top-ranking answers learned by our approach is much higher than that of plurality voting (In one question, the plurality answer is supported by 74 respondents but the correct answer is only supported by 3 respondents. Our approach ranks the correct answer the highest without any prior); b) our model has a high goodness-of-fit, especially for the questions where our top-ranking answer is correct. To the best of our knowledge, we are the first to propose a thinking hierarchy model with empirical validations in the general problem-solving scenarios; and the first to propose a practical open-response-based crowdsourcing approach that beats plurality voting without any prior.

Multi-round competitions often double or triple the points awarded in the final round, calling it a bonus, to maximize spectators’ excitement. In a two-player competition with n rounds, we aim to derive the optimal bonus size to maximize the audience’s overall expected surprise (as defined in [7]). We model the audience’s prior belief over the two players’ ability levels as a beta distribution. Using a novel analysis that clarifies and simplifies the computation, we find that the optimal bonus depends greatly upon the prior belief and obtain solutions of various forms for both the case of a finite number of rounds and the asymptotic case. In an interesting special case, we show that the optimal bonus approximately and asymptotically equals to the “expected lead”, the number of points the weaker player will need to come back in expectation. Moreover, we observe that priors with a higher skewness lead to a higher optimal bonus size, and in the symmetric case, priors with a higher uncertainty also lead to a higher optimal bonus size. This matches our intuition since a highly asymmetric prior leads to a high “expected lead”, and a highly uncertain symmetric prior often leads to a lopsided game, which again benefits from a larger bonus.

Information flow measures, over the duration of a game, the audience’s belief of who will win, and thus can reflect the amount of surprise in a game. To quantify the relationship between information flow and audiences' perceived quality, we conduct a case study where subjects watch one of the world’s biggest esports events, LOL S10. In addition to eliciting information flow, we also ask subjects to report their rating for each game. We find that the amount of surprise in the end of the game plays a dominant role in predicting the rating. This suggests the importance of incorporating when the surprise occurs, in addition to the amount of surprise, in perceived quality models. For content providers, it implies that everything else being equal, it is better for twists to be more likely to happen toward the end of a show rather than uniformly throughout.

We study learning statistical properties from strategic agents with private information. In this problem, agents must be incentivized to truthfully reveal their information even when it cannot be directly verified. Moreover, the information reported by the agents must be aggregated into a statistical estimate. We study two fundamental statistical properties: estimating the mean of an unknown Gaussian, and linear regression with Gaussian error. The information of each agent is one point in a Euclidean space.Our main results are two mechanisms for each of these problems which optimally aggregate the information of agents in the truth-telling equilibrium:• A minimal (non-revelation) mechanism for large populations — agents only need to report one value, but that value need not be their point.• A mechanism for small populations that is non-minimal — agents need to answer more than one question.These mechanisms are “informed truthful” mechanisms where reporting unaltered data (truth-telling) 1) forms a strict Bayesian Nash equilibrium and 2) has strictly higher welfare than any oblivious equilibrium where agents' strategies are independent of their private signals. We also show a minimal revelation mechanism (each agent only reports her signal) for a restricted setting and use an impossibility result to prove the necessity of this restriction.We build upon the peer prediction literature in the single-question setting; however, most previous work in this area focuses on discrete signals, whereas our setting is inherently continuous, and we further simplify the agents' reports.

In the setting where participants are asked multiple similar possibly subjective multi-choice questions (e.g. Do you like Panda Express? Y/N; do you like Chick-fil-A? Y/N), a series of peer prediction mechanisms are designed to incentivize honest reports and some of them achieve dominantly truthfulness: truth-telling is a dominant strategy and strictly dominate other “non-permutation strategy” with some mild conditions. However, a major issue hinders the practical usage of those mechanisms: they require the participants to perform an infinite number of tasks. When the participants perform a finite number of tasks, these mechanisms only achieve approximated dominant truthfulness. The existence of a dominantly truthful multi-task peer prediction mechanism that only requires a finite number of tasks remains to be an open question that may have a negative result, even with full prior knowledge. This paper answers this open question by proposing a new mechanism, Determinant based Mutual Information Mechanism (DMI-Mechanism), that is dominantly truthful when the number of tasks is ≥ 2C. C is the number of choices for each question (C = 2 for binary-choice questions). DMI-Mechanism also pays truth-telling higher than any strategy profile and strictly higher than uninformative strategy profiles (informed truthfulness). In addition to the truthfulness properties, DMI-Mechanism is also easy to implement since it does not require any prior knowledge (detail-free) and only requires ≥ 2 participants. The core of DMI-Mechanism is a novel information measure, Determinant based Mutual Information (DMI). DMI generalizes Shannon's mutual information and the square of DMI has a simple unbiased estimator. In addition to incentivizing honest reports, DMI-Mechanism can also be transferred into an information evaluation rule that identifies high-quality information without verification when there are ≥ 3 participants. To the best of our knowledge, DMI-Mechanism is both the first detail-free informed-truthful mechanism and the first dominantly truthful mechanism that works for a finite number of tasks, not to say a small constant number of tasks.

In this paper, we propose a Deep Metric Learning (DML) approach that supports soft labels. DML seeks to learn representations that encode the similarity between examples through deep neural networks. DML generally presupposes that data can be divided into discrete classes using hard labels. However, some tasks, such as our exemplary domain of speech emotion recognition (SER), work with inherently subjective data, data for which it may not be possible to identify a single hard label. We propose a family of loss functions, fSimilarity Preservation Loss (f-SPL), based on the dual form of f-divergence for DML with soft labels. We show that the minimizer of f-SPL preserves the pairwise label similarities in the learned feature embeddings. We demonstrate the efficacy of the proposed loss function on the task of cross-corpus SER with soft labels. Our approach, which combines f-SPL and classification loss, significantly outperforms a baseline SER system with the same structure but trained with only classification loss in most experiments. We show that the presented techniques are more robust to over-training and can learn an embedding space in which the similarity between examples is meaningful.

Accurately annotating large scale dataset is notoriously expensive both in time and in money. Although acquiring low-quality-annotated dataset can be much cheaper, it often badly damages the performance of trained models when using such dataset without particular treatment. Various methods have been proposed for learning with noisy labels. However, most methods only handle limited kinds of noise patterns, require auxiliary information or steps (e.g., knowing or estimating the noise transition matrix), or lack theoretical justification. In this paper, we propose a novel information-theoretic loss function, LDMI, for training deep neural networks robust to label noise. The core of LDMI is a generalized version of mutual information, termed Determinant based Mutual Information (DMI), which is not only information-monotone but also relatively invariant. To the best of our knowledge, LDMI is the first loss function that is provably robust to instance-independent label noise, regardless of noise pattern, and it can be applied to any existing classification neural networks straightforwardly without any auxiliary information. In addition to theoretical justification, we also empirically show that using LDMI outperforms all other counterparts in the classification task on both image dataset and natural language dataset include Fashion-MNIST, CIFAR-10, Dogs vs. Cats, MR with a variety of synthesized noise patterns and noise amounts, as well as a real-world dataset Clothing1M.

In the setting where information cannot be verified, we propose a simple yet powerful information theoretical framework---the Mutual Information Paradigm---for information elicitation mechanisms. Our framework pays every agent a measure of mutual information between her signal and a peer's signal. We require that the mutual information measurement has the key property that any "data processing" on the two random variables will decrease the mutual information between them. We identify such information measures that generalize Shannon mutual information. Our Mutual Information Paradigm overcomes the two main challenges in information elicitation without verification: (1) how to incentivize effort and avoid agents colluding to report random or identical responses (2) how to motivate agents who believe they are in the minority to report truthfully. Aided by the information measures we found, (1) we use the paradigm to design a family of novel mechanisms where truth-telling is a dominant strategy and any other strategy will decrease every agent's expected payment (in the multi-question, detail free, minimal setting where the number of questions is large); (2) we show the versatility of our framework by providing a unified theoretical understanding of existing mechanisms---Peer Prediction [Miller 2005], Bayesian Truth Serum [Prelec 2004], and Dasgupta and Ghosh [2013]---by mapping them into our framework such that theoretical results of those existing mechanisms can be reconstructed easily. We also give an impossibility result which illustrates, in a certain sense, the the optimality of our framework.

Prediction markets provide a unique and compelling way to sell and aggregate information, yet a good understanding of optimal strategies for agents participating in such markets remains elusive. To model this complex setting, prior work proposes a three stages game called the Alice Bob Alice (A-B-A) game - Alice participates in the market first, then Bob joins, and then Alice has a chance to participate again. While prior work has made progress in classifying the optimal strategy for certain interesting edge cases, it remained an open question to calculate Alice's best strategy in the A-B-A game for a general information structure. In this paper, we analyze the A-B-A game for a general information structure and (1) show a "revelation-principle" style result: it is enough for Alice to use her private signal space as her announced signal space, that is, Alice cannot gain more by revealing her information more "finely"; (2) provide a FPTAS to compute the optimal information revelation strategy with additive error when Alice's information is a signal from a constant-sized set; (3) show that sometimes it is better for Alice to reveal partial information in the first stage even if Alice's information is a single binary bit.

We build a natural connection between the learning problem, co-training, and forecast elicitation without verification (related to peer-prediction) and address them simultaneously using the same information theoretic approach. In co-training/multiview learning, the goal is to aggregate two views of data into a prediction for a latent label. We show how to optimally combine two views of data by reducing the problem to an optimization problem. Our work gives a unified and rigorous approach to the general setting. In forecast elicitation without verification we seek to design a mechanism that elicits high quality forecasts from agents in the setting where the mechanism does not have access to the ground truth. By assuming the agents' information is independent conditioning on the outcome, we propose mechanisms where truth-telling is a strict equilibrium for both the single-task and multi-task settings. Our multi-task mechanism additionally has the property that the truth-telling equilibrium pays better than any other strategy profile and strictly better than any other "non-permutation" strategy profile.

A central question of crowdsourcing is how to elicit expertise from agents. This is even more difficult when answers cannot be directly verified. A key challenge is that sophisticated agents may strategically withhold effort or information when they believe their payoff will be based upon comparison with other agents whose reports will likely omit this information due to lack of effort or expertise. Our work defines a natural model for this setting based on the assumption that more sophisticated agents know the beliefs of less sophisticated agents. We then provide a mechanism design framework for this setting. From this framework, we design several novel mechanisms, for both the single and multiple tasks settings, that (1) encourage agents to invest effort and provide their information honestly; (2) output a correct "hierarchy" of the information when agents are rational.