United States

Recent literature uses language to build foundation models for audio. These Audio-Language Models (ALMs) are trained on a vast number of audio-text pairs and show remarkable performance in tasks including Text-to-Audio Retrieval, Captioning, and Question Answering. However, their ability to engage in more complex open-ended tasks, like Interactive Question-Answering, requires proficiency in logical reasoning-- a skill not yet benchmarked. We introduce the novel task of Audio Entailment to evaluate an ALM&#39;s deductive reasoning ability. This task assesses whether a text description (hypothesis) of audio content can be deduced from an audio recording (premise), with potential conclusions being entailment, neutral, or contradiction, depending on the sufficiency of the evidence. We create two datasets for this task with audio recordings sourced from two audio captioning datasets-AudioCaps and Clotho-and hypotheses generated using Large Language Models (LLMs). We benchmark state-of-the-art ALMs and find deficiencies in logical reasoning with both zero-shot and linear probe evaluations. Finally, we propose &quot;caption-before-reason&quot;, an intermediate step of captioning that improves the Zero-Shot and linear-probe performance of ALMs by an absolute 6% and 3%, respectively.

AAAI 2025

Audio Entailment: Assessing Deductive Reasoning for Audio Understanding

speech signal processing

snlp

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Please register!

NAACL 2025

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Recent literature uses language to build foundation models for audio. These Audio-Language Models (ALMs) are trained on a vast number of audio-text pairs and show remarkable performance in tasks including Text-to-Audio Retrieval, Captioning, and Question Answering. However, their ability to engage in more complex open-ended tasks, like Interactive Question-Answering, requires proficiency in logical reasoning-- a skill not yet benchmarked. We introduce the novel task of Audio Entailment to evaluate an ALM's deductive reasoning ability. This task assesses whether a text description (hypothesis) of audio content can be deduced from an audio recording (premise), with potential conclusions being entailment, neutral, or contradiction, depending on the sufficiency of the evidence. We create two datasets for this task with audio recordings sourced from two audio captioning datasets-AudioCaps and Clotho-and hypotheses generated using Large Language Models (LLMs). We benchmark state-of-the-art ALMs and find deficiencies in logical reasoning with both zero-shot and linear probe evaluations. Finally, we propose "caption-before-reason", an intermediate step of captioning that improves the Zero-Shot and linear-probe performance of ALMs by an absolute 6% and 3%, respectively.

technical paper

We are pleased to announce the Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI-25), which will be held in Philadelphia, Pennsylvania at the Pennsylvania Convention Center from February 25 to March 4, 2025.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-25 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

### [Invited Speakers](https://aaai.org/conference/aaai/aaai-25/aaai-25-invited-speakers/)

Register [here](https://aaai.org/conference/aaai/aaai-25/registration/)

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-25 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.



Federated Learning (FL) is a distributed machine learning (ML) paradigm, in which multiple clients collaboratively train ML models without centralizing their local data. Similar to conventional ML pipelines, the client local optimization and server aggregation procedure in FL are sensitive to the hyperparameter (HP) selection. Despite extensive research on tuning HPs for centralized ML, these methods yield suboptimal results when employed in FL. This is mainly because their "training-after-tuning" framework is unsuitable for FL with limited client computation power. While some approaches have been proposed for HP-Tuning in FL, they are limited to the HPs for client local updates. In this work, we propose a novel HP-tuning algorithm, called Federated Population-based Hyperparameter Tuning (FedPop), to address this vital yet challenging problem. FedPop employs population-based evolutionary algorithms to optimize the HPs, which accommodates various HP types at both the client and server sides. Compared with prior tuning methods, FedPop employs an online  "tuning-while-training" framework, offering computational efficiency and enabling the exploration of a broader HP search space. Our empirical validation on the common FL benchmarks and complex real-world FL datasets, including full-sized Non-IID ImageNet-1K, demonstrates the effectiveness of the proposed method, which substantially outperforms the concurrent state-of-the-art HP-tuning methods in FL.

FedPop: Federated Population-based Hyperparameter Tuning

Personalized Federated Learning (PFL) is widely used in IoT applications to manage large-scale, non-iid client data while ensuring data privacy. However, the diverse edge devices owned by clients may impose varying degrees of resource constraints, leading to computation and communication bottlenecks for PFL. To tackle this challenge, Federated Dropout has emerged as a popular strategy, where only a subset of the global model, known as a sub-model, is trained on a client's device, thereby reducing computation and communication overheads. Nevertheless, the dropout-based model-pruning strategy may introduce bias, especially towards non-iid local data. This bias can result in performance degradation when biased sub-models absorb highly divergent parameters from other clients. In response, we propose federated learning with stochastic parameter update (FedSPU). Unlike dropout, which tailors the global model to small-size local sub-models, FedSPU maintains the full model architecture on each device but randomly freezes a certain percentage of neurons in the local model during training while updating the remaining neurons. This approach ensures that a portion of the local model remains personalized, thereby enhancing the model's robustness against biased parameters from other clients. Experimental results demonstrate that FedSPU outperforms federated dropout by an average of 7.57% in terms of accuracy. Furthermore, an introduced early stopping scheme leads to a significant reduction in training time by 24.8% to 70.4% while maintaining high accuracy.

FedSPU: Personalized Federated Learning for Resource-Constrained Devices with Stochastic Parameter Update

We study the problem of computing fair divisions of a set of indivisible goods among agents with additive valuations. For the past many decades, the literature has explored various notions of fairness, that can be primarily seen as either having envy-based or share-based lens. For the discrete setting of resource-allocation problems, envy-free up to any good (EFX) and maximin share (MMS) are widely considered as the flag-bearers of fairness notions in the above two categories, thereby capturing different aspects of fairness herein. Due to lack of existence results of these notions and the fact that a good approximation of EFX or MMS does not imply particularly strong guarantees of the other, it becomes important to understand the compatibility of EFX and MMS allocations with one another.

In this work, we identify a novel way to simultaneously achieve MMS guarantees with EFX/EF1 notions of fairness, while beating the best known approximation factors (Chaudhury et al. 2021; Amanatidis, Markakis, and Ntokos 2020). Our main contribution is to constructively prove the existence of (i) a partial allocation that is both $2/3$-MMS and EFX, and (ii) a complete allocation that is both $2/3$-MMS and EF1. Our algorithms run in pseudo-polynomial time if the approximation factor for MMS is relaxed to $2/3$-$\varepsilon$ for any constant \varepsilon>0 and in polynomial time if, in addition, the EFX (or EF1) guarantee is relaxed to $(1-\delta)$-EFX (or $(1-\delta)$-EF1) for any constant $\delta>0$. In particular, we improve from the best approximation factor known prior to our work, which computes partial allocations that are $1/2$-MMS and EFX in pseudo-polynomial time (Chaudhury et al. 2021).

Achieving Maximin Share and EFX/EF1 Guarantees Simultaneously

We study the fundamental problem of fairly dividing a set of indivisible items among agents with (general) monotone valuations. The notion of envy-freeness up to any item (EFX) is considered to be one of the most fascinating fairness concepts in this line of work. Unfortunately, despite significant efforts, existence of EFX allocations is a major open problem in fair division, thereby making the study of approximations and relaxations of EFX a natural line of research. Recently, Caragiannis et al. [2023] introduced a promising relaxation of EFX, called epistemic EFX (EEFX). We say an allocation to be EEFX if, for every agent, it is possible to shuffle the items in the remaining bundles so that she becomes “EFX-satisfied”. Caragiannis et al. [2023] prove existence and polynomial-time computability of EEFX allocations for additive valuations. A natural question asks what happens when we consider valuations more general than additive?

We address this important open question and answer it affirmatively by establishing the existence of EEFX allocations for an arbitrary number of agents with general monotone valuations. To the best of our knowledge, EEFX is the only known relaxation of EFX (beside EF1) to have such strong existential guarantees. Furthermore, we complement our existential result by proving computational and information-theoretic lower bounds. We prove that even for an arbitrary number of (more than one) agents with identical submodular valuations, it is PLS-hard to compute EEFX allocations and it requires exponentially-many value queries to do so.

Epistemic EFX Allocations Exist for Monotone Valuations

The maximin optimisation problem, inspired by von Neumann’s work (von Neumann 1928) and widely applied in adversarial optimisation, has become a key research area in machine learning. Gradient Descent Ascent (GDA) is a common method for solving these problems but requires the payoff function to be differentiable, making it unsuitable for discrete or binary functions that often occur in game-theoretical scenarios. Co-evolutionary algorithms (CoEAs), which are derivative-free, offer an alternative to these problems. However, the theoretical understanding of CoEAs is still limited.

This paper provides the first rigorous runtime analysis of CoEAs with pairwise dominance on binary two-player zero-sum games (or maximin problems), specifically focusing on the DIAGONAL game. The mathematical analysis rigorously shows that the PDCoEA can efficiently find the optimum in polynomial runtime with high probability under low mutation rates and large population sizes. Empirical evidence also identifies an error threshold where higher mutation rates lead to inefficiency. In contrast, single-pair-individual algorithms, i.e.RLS-PD and (1+1)-CoEAs, fail to find the optimum in polynomial time. These findings highlight the importance of pairwise dominance, low mutation rates, and large populations in maintaining the “co-evolutionary arms race”.

Towards Runtime Analysis of Population-Based Co-evolutionary Algorithms on Sparse Binary Zero-Sum Game

Current neural networks often employ multi-domain-learning or attribute-injecting mechanisms to incorporate non-independent and identically distributed (non-IID) information for text understanding tasks by capturing individual characteristics and the relationships among samples. However, the extent of the impact of non-IID information and how these methods affect pre-trained language models (PLMs) remains unclear. This study revisits the assumption that non-IID information enhances PLMs to achieve performance improvements from a Bayesian perspective, which unearths and integrates non-IID and IID features. Furthermore, we proposed a multi-attribute multi-grained framework for PLM adaptations (M2A), which combines multi-attribute and multi-grained views to mitigate uncertainty in a lightweight manner. We evaluate M2A through prevalent text-understanding datasets and demonstrate its superior performance, mainly when data are implicitly non-IID, and PLMs scale larger. Our code is publicly available at https://anonymous.4open.science/r/M2A-62EA/.

Multi-attribute Multi-grained Adaptation of Pre-trained Language Models for Text Understanding from Bayesian Perspective

Responsibility plays a key role in the development and deployment of trustworthy autonomous systems. In this paper, we focus on the problem of strategic reasoning in probabilistic multi-agent systems with responsibility-aware agents. We introduce the logic PATL+R, a variant of Probabilistic Alternating-time Temporal Logic. The novelty of PATL+R lies in its incorporation of modalities for causal responsibility, providing a framework for responsibility-aware multi-agent strategic reasoning. We present an approach to synthesise joint strategies that satisfy an outcome specified in PATL+R, while optimising the share of expected causal responsibility and reward. This provides a notion of balanced distribution of responsibility and reward gain among agents. To this end, we utilise the Nash equilibrium as the solution concept for our strategic reasoning problem and demonstrate how to compute responsibility-aware Nash equilibrium strategies via a reduction to parametric model checking of concurrent stochastic multi-player games.

Responsibility-aware Strategic Reasoning in Probabilistic Multi-Agent Systems

Multi-Robot Coverage problems have been extensively studied in robotics, planning and multi-agent systems. In this work, we consider the coverage problem when there are constraints on the proximity (e.g., maximum distance between the agents, or a blue agent must be adjacent to a red agent) and the movement (e.g., terrain traversability and material load capacity) of the robots.
Such constraints naturally arise in many real-world applications, e.g. in search-and-rescue and maintenance operations. Given such a setting, the goal is to compute a covering tour of the graph with a minimum number of steps, and that adheres to the proximity and movement constraints.   
For this problem, our contributions are four: (i) a formal formulation of the problem, (ii) an exact algorithm that is FPT in parameters $\|\mathcal{F}\|$, $d$ and $\textsf{tw}$ - the set of robot formations that encode the proximity constraints, the maximum nodes degree, and the tree-width of the graph, respectively, (iii) for the case that the graph is a tree: a PTAS approximation scheme, that given an $\varepsilon$ produces a tour that is within a $1+\varepsilon\cdot \textsf{error}(\|\mathcal{F}\|,d))$ of the optimal one, and the computation runs in time $\textsf{poly}(n) \cdot h(\frac{1}{\varepsilon},\mathcal{F})$.
(iv) for the case that the graph is a tree, with $k=3$ robots, and the constraint is that all agents are connected: a PTAS scheme with multiplicative approximation error of $1+\mathcal{O}(\varepsilon)$, independent of $d$.

Heterogeneous Multi-Robot Graph Coverage with Proximity and Movement Constraints

The group recommendation (GR) aims to suggest items for a group of users in social networks. Existing work typically considers individual preferences as the sole factor in aggregating group preferences. Actually, social influence is also an important factor in modeling users' contributions to the final group decision. However, existing methods either neglect the social influence of individual members or bundle preferences and social influence together as a unified representation. As a result, these models emphasize the preferences of the majority within the group rather than the actual interaction items, which we refer to as the preference bias issue in GR. Moreover, the self-supervised learning (SSL) strategies they designed to address the issue of group data sparsity fail to account for users' contextual social weights when regulating group representations, leading to suboptimal results. To tackle these issues, we propose a novel model based on Disentangled Modeling of Preferences and Social Influence for Group Recommendation (DisRec). Concretely, we first design a user-level disentangling network to disentangle the preferences and social influence of group members with separate embedding propagation schemes based on (hyper)graph convolution networks. We then introduce a social-based contrastive learning strategy, selectively excluding user nodes based on their social importance to enhance group representations and alleviate the group-level data sparsity issue. The experimental results demonstrate that our model significantly outperforms state-of-the-art methods on two real-world datasets.

Disentangled Modeling of Preferences and Social Influence for Group Recommendation

Multiple signal modalities, such as vision and sounds, are naturally present in real-world phenomena. Recently, there has been growing interest in learning generative models, in particular variational autoencoder (VAE), to for multimodal representation learning especially in the case of missing modalities. The primary goal of these models is to learn a modality-invariant and modality-specific representation that characterizes information across multiple modalities. 
Previous attempts at multimodal VAEs approach this mainly through the lens of experts, aggregating unimodal inference distributions with a product of experts (PoE), a mixture of experts (MoE), or a combination of both. In this paper, we provide an alternative generic and theoretical formulation of multimodal VAE through the lens of barycenter. We first show that PoE and MoE are specific instances of barycenters, derived by minimizing the asymmetric weighted KL divergence to unimodal inference distributions. Our novel formulation extends these two barycenters to a more flexible choice by considering different types of divergences.
In particular, we explore the Wasserstein barycenter defined by the 2-Wasserstein distance, which better preserves the geometry of unimodal distributions by capturing both modality-specific and modality-invariant representations compared to KL divergence. Empirical studies on three multimodal benchmarks demonstrated the effectiveness of the proposed method.

Downloads

Next from AAAI 2025

FedPop: Federated Population-based Hyperparameter Tuning

Similar lecture

PAT: Parameter-Free Audio-Text Aligner to Boost Zero-Shot Audio Classification