Transformer-based Language Models have become ubiquitous in Natural Language Processing (NLP) due to their impressive performance on various tasks. However, expensive training as well as inference remains a significant impediment to their widespread applicability. While enforcing sparsity at various levels of the model architecture has found promise in addressing scaling and efficiency issues, there remains a disconnect between how sparsity affects network topology. Inspired by brain neuronal networks, we explore sparsity approaches through the lens of network topology. Specifically, we exploit mechanisms seen in biological networks, such as preferential attachment and redundant synapse pruning, and show that principled, model-agnostic sparsity approaches are performant and efficient across diverse NLP tasks, spanning both classification (such as natural language inference) and generation (summarization, machine translation), despite our sole objective not being optimizing performance. NeuroPrune is competitive with (or sometimes superior to) baselines on performance and can be up to $10$x faster in terms of training time for a given level of sparsity, simultaneously exhibiting measurable improvements in inference time in many cases.

Collecting time series data spatially distributed in many locations is often important for analyzing climate change and its impacts on ecosystems. However, comprehensive spatial data collection is not always feasible, requiring us to predict climate variables at some locations. This study focuses on a prediction of climatic elements, specifically near-surface temperature and pressure, at a target location apart from a data observation point. Our approach uses two prediction methods: reservoir computing (RC), known as a machine learning framework with low computational requirements, and vector autoregression models (VAR), recognized as a statistical method for analyzing time series data. Our results show that the accuracy of the predictions degrades with the distance between the observation and target locations. We quantitatively estimate the distance in which effective predictions are possible. We also find that in the context of climate data, a geographical distance is associated with data correlation, and a strong data correlation significantly improves the prediction accuracy with RC. In particular, RC outperforms VAR in predicting highly correlated data within the predictive range. These findings suggest that machine learning-based methods can be used more effectively to predict climatic elements in remote locations by assessing the distance to them from the data observation point in advance. Our study on low-cost and accurate prediction of climate variables has significant value for climate change strategies.

The conference featured lots of robots (including one that dispenses wine), but what I liked most of all was how it managed to convene people working in AI from around the globe, featuring speakers from China, the Middle East, and Africa too, such as Pelonomi Moiloa, the CEO of Lelapa AI, a startup building AI for African languages. AI can be very US-centric and male dominated, and any effort to make the conversation more global and diverse is laudable. But honestly, I didn't leave the conference feeling confident AI was going to play a meaningful role in advancing any of the UN goals. In fact, the most interesting speeches were about how AI is doing the opposite. Sage Lenier, a climate activist, talked about how we must not let AI accelerate environmental destruction.

African languages are not well-represented in natural language processing (NLP). This is in large part due to a lack of resources for training models. Henok Biadglign Ademtew and Mikiyas Girma Birbo have created an Amharic, Ge'ez, and English parallel dataset to help advance research into low-resource languages. We spoke to Henok about this project, the creation of the dataset, and some of the challenges faced. Most of the languages in Africa are very low-resourced, and not much text data is available.

We study the iteration complexity of decentralized learning of approximate correlated equilibria in incomplete information games. On the negative side, we prove that in $\mathit{extensive}$-$\mathit{form}$ $\mathit{games}$, assuming $\mathsf{PPAD} \not\subset \mathsf{TIME}(n^{\mathsf{polylog}(n)})$, any polynomial-time learning algorithms must take at least $2^{\log_2^{1-o(1)}(|\mathcal{I}|)}$ iterations to converge to the set of $\epsilon$-approximate correlated equilibrium, where $|\mathcal{I}|$ is the number of nodes in the game and $\epsilon > 0$ is an absolute constant. This nearly matches, up to the $o(1)$ term, the algorithms of [PR'24, DDFG'24] for learning $\epsilon$-approximate correlated equilibrium, and resolves an open question of Anagnostides, Kalavasis, Sandholm, and Zampetakis [AKSZ'24]. Our lower bound holds even for the easier solution concept of $\epsilon$-approximate $\mathit{coarse}$ correlated equilibrium On the positive side, we give uncoupled dynamics that reach $\epsilon$-approximate correlated equilibria of a $\mathit{Bayesian}$ $\mathit{game}$ in polylogarithmic iterations, without any dependence of the number of types. This demonstrates a separation between Bayesian games and extensive-form games.

In practice, training using federated learning can be orders of magnitude slower than standard centralized training. This severely limits the amount of experimentation and tuning that can be done, making it challenging to obtain good performance on a given task. Server-side proxy data can be used to run training simulations, for instance for hyperparameter tuning. This can greatly speed up the training pipeline by reducing the number of tuning runs to be performed overall on the true clients. However, it is challenging to ensure that these simulations accurately reflect the dynamics of the real federated training. In particular, the proxy data used for simulations often comes as a single centralized dataset without a partition into distinct clients, and partitioning this data in a naive way can lead to simulations that poorly reflect real federated training. In this paper we address the challenge of how to partition centralized data in a way that reflects the statistical heterogeneity of the true federated clients. We propose a fully federated, theoretically justified, algorithm that efficiently learns the distribution of the true clients and observe improved server-side simulations when using the inferred distribution to create simulated clients from the centralized data.

Human-wildlife conflict challenges conservation worldwide, which requires innovative management solutions. We developed a prototype Agent-Based Model (ABM) to simulate interactions between humans and solitary bull Asian elephants in the Periyar-Agasthyamalai complex of the Western Ghats in Kerala, India. The main challenges were the complex behavior of elephants and insufficient movement data from the region. Using literature, expert insights, and field surveys, we created a prototype behavior model that incorporates crop habituation, thermoregulation, and aggression. We designed a four-step calibration method to adapt relocation data from radio-tagged elephants in Indonesia to model elephant movements in the model domain. The ABM's structure, including the assumptions, submodels, and data usage are detailed following the Overview, Design concepts, Details protocol. The ABM simulates various food availability scenarios to study elephant behavior and environmental impact on space use and conflict patterns. The results indicate that the wet months increase conflict and thermoregulation significantly influences elephant movements and crop raiding. Starvation and crop habituation intensify these patterns. This prototype ABM is an initial model that offers information on the development of a decision support system in wildlife management and will be further enhanced with layers of complexity and subtlety across various dimensions. Access the ABM at https://github.com/quest-lab-iisc/abm-elephant-project.

Building machines with commonsense has been a longstanding challenge in NLP due to the reporting bias of commonsense rules and the exposure bias of rule-based commonsense reasoning. In contrast, humans convey and pass down commonsense implicitly through stories. This paper investigates the inherent commonsense ability of large language models (LLMs) expressed through storytelling. We systematically investigate and compare stories and rules for retrieving and leveraging commonsense in LLMs. Experimental results on 28 commonsense QA datasets show that stories outperform rules as the expression for retrieving commonsense from LLMs, exhibiting higher generation confidence and commonsense accuracy. Moreover, stories are the more effective commonsense expression for answering questions regarding daily events, while rules are more effective for scientific questions. This aligns with the reporting bias of commonsense in text corpora. We further show that the correctness and relevance of commonsense stories can be further improved via iterative self-supervised fine-tuning. These findings emphasize the importance of using appropriate language to express, retrieve, and leverage commonsense for LLMs, highlighting a promising direction for better exploiting their commonsense abilities.

We study the impact of the batch size $n_b$ on the iteration time $T$ of training two-layer neural networks with one-pass stochastic gradient descent (SGD) on multi-index target functions of isotropic covariates. We characterize the optimal batch size minimizing the iteration time as a function of the hardness of the target, as characterized by the information exponents. We show that performing gradient updates with large batches $n_b \lesssim d^{\frac{\ell}{2}}$ minimizes the training time without changing the total sample complexity, where $\ell$ is the information exponent of the target to be learned \citep{arous2021online} and $d$ is the input dimension. However, larger batch sizes than $n_b \gg d^{\frac{\ell}{2}}$ are detrimental for improving the time complexity of SGD. We provably overcome this fundamental limitation via a different training protocol, \textit{Correlation loss SGD}, which suppresses the auto-correlation terms in the loss function. We show that one can track the training progress by a system of low-dimensional ordinary differential equations (ODEs). Finally, we validate our theoretical results with numerical experiments.

Exposure assessment is fundamental to air pollution cohort studies. The objective is to predict air pollution exposures for study subjects at locations without data in order to optimize our ability to learn about health effects of air pollution. In addition to generating accurate predictions to minimize exposure measurement error, understanding the mechanism captured by the model is another crucial aspect that may not always be straightforward due to the complex nature of machine learning methods, as well as the lack of unifying notions of variable importance. This is further complicated in air pollution modeling by the presence of spatial correlation. We tackle these challenges in two datasets: sulfur (S) from regulatory United States national PM2.5 sub-species data and ultrafine particles (UFP) from a new Seattle-area traffic-related air pollution dataset. Our key contribution is a leave-one-out approach for variable importance that leads to interpretable and comparable measures for a broad class of models with separable mean and covariance components. We illustrate our approach with several spatial machine learning models, and it clearly highlights the difference in model mechanisms, even for those producing similar predictions. We leverage insights from this variable importance measure to assess the relative utilities of two exposure models for S and UFP that have similar out-of-sample prediction accuracies but appear to draw on different types of spatial information to make predictions.