The prediction of streamflows and other environmental variables in unmonitored basins is a grand challenge in hydrology. Recent machine learning (ML) models can harness vast datasets for accurate predictions at large spatial scales. However, there are open questions regarding model design and data needed for inputs and training to improve performance. This study explores these questions while demonstrating the ability of deep learning models to make accurate stream temperature predictions in unmonitored basins across the conterminous United States. First, we compare top-down models that utilize data from a large number of basins with bottom-up methods that transfer ML models built on local sites, reflecting traditional regionalization techniques. We also evaluate an intermediary grouped modeling approach that categorizes sites based on regional co-location or similarity of catchment characteristics. Second, we evaluate trade-offs between model complexity, prediction accuracy, and applicability for more target locations by systematically removing inputs. We then examine model performance when additional training data becomes available due to reductions in input requirements. Our results suggest that top-down models significantly outperform bottom-up and grouped models. Moreover, it is possible to get acceptable accuracy by reducing both dynamic and static inputs enabling predictions for more sites with lower model complexity and computational needs. From detailed error analysis, we determined that the models are more accurate for sites primarily controlled by air temperatures compared to locations impacted by groundwater and dams. By addressing these questions, this research offers a comprehensive perspective on optimizing ML model design for accurate predictions in unmonitored regions.

With the capability to write convincing and fluent natural language and generate code, Foundation Models present dual-use concerns broadly and within the cyber domain specifically. Generative AI has already begun to impact cyberspace through a broad illicit marketplace for assisting malware development and social engineering attacks through hundreds of malicious-AI-as-a-services tools. More alarming is that recent research has shown the potential for these advanced models to inform or independently execute offensive cyberspace operations. However, these previous investigations primarily focused on the threats posed by proprietary models due to the until recent lack of strong open-weight model and additionally leave the impacts of network defenses or potential countermeasures unexplored. Critically, understanding the aptitude of downloadable models to function as offensive cyber agents is vital given that they are far more difficult to govern and prevent their misuse. As such, this work evaluates several state-of-the-art FMs on their ability to compromise machines in an isolated network and investigates defensive mechanisms to defeat such AI-powered attacks. Using target machines from a commercial provider, the most recently released downloadable models are found to be on par with a leading proprietary model at conducting simple cyber attacks with common hacking tools against known vulnerabilities. To mitigate such LLM-powered threats, defensive prompt injection (DPI) payloads for disrupting the malicious cyber agent's workflow are demonstrated to be effective. From these results, the implications for AI safety and governance with respect to cybersecurity is analyzed.

We propose a new model for multi-token prediction in transformers, aiming to enhance sampling efficiency without compromising accuracy. Motivated by recent work that predicts the probabilities of subsequent tokens using multiple heads, we connect this approach to rank-$1$ canonical tensor decomposition. By generalizing it to a rank-$r$ canonical probability decomposition, we develop an improved model that predicts multiple tokens simultaneously. This model can also be interpreted as a mixture of experts, allowing us to leverage successful techniques from that domain for efficient and robust training. Importantly, the overall overhead for training and sampling remains low. Our method demonstrates significant improvements in inference speed for both text and code generation tasks, proving particularly beneficial within the self-speculative decoding paradigm. It maintains its effectiveness across various model sizes and training epochs, highlighting its robustness and scalability.

The quest for simplification in physics drives the exploration of concise mathematical representations for complex systems. This Dissertation focuses on the concept of dimensionality reduction as a means to obtain low-dimensional descriptions from high-dimensional data, facilitating comprehension and analysis. We address the challenges posed by real-world data that defy conventional assumptions, such as complex interactions within neural systems or high-dimensional dynamical systems. Leveraging insights from both theoretical physics and machine learning, this work unifies diverse reduction methods under a comprehensive framework, the Deep Variational Multivariate Information Bottleneck. This framework enables the design of tailored reduction algorithms based on specific research questions. We explore and assert the efficacy of simultaneous reduction approaches over their independent reduction counterparts, demonstrating their superiority in capturing covariation between multiple modalities, while requiring less data. We also introduced novel techniques, such as the Deep Variational Symmetric Information Bottleneck, for general nonlinear simultaneous reduction. We show that the same principle of simultaneous reduction is the key to efficient estimation of mutual information. We show that our new method is able to discover the coordinates of high-dimensional observations of dynamical systems. Through analytical investigations and empirical validations, we shed light on the intricacies of dimensionality reduction methods, paving the way for enhanced data analysis across various domains. We underscore the potential of these methodologies to extract meaningful insights from complex datasets, driving advancements in fundamental research and applied sciences. As these methods evolve, they promise to deepen our understanding of complex systems and inform more effective data analysis strategies.

Inverse dynamics (ID), which aims at reproducing the driven torques from human kinematic observations, has been a critical tool for gait analysis. However, it is hindered from wider application to general motion due to its limited scalability. Conventional optimization-based ID requires expensive laboratory setups, restricting its availability. To alleviate this problem, we propose to exploit the recently progressive human motion imitation algorithms to learn human inverse dynamics in a data-driven manner. The key insight is that the human ID knowledge is implicitly possessed by motion imitators, though not directly applicable. In light of this, we devise an efficient data collection pipeline with state-of-the-art motion imitation algorithms and physics simulators, resulting in a large-scale human inverse dynamics benchmark as Imitated Dynamics (ImDy). ImDy contains over 150 hours of motion with joint torque and full-body ground reaction force data. With ImDy, we train a data-driven human inverse dynamics solver ImDyS(olver) in a fully supervised manner, which conducts ID and ground reaction force estimation simultaneously. Experiments on ImDy and real-world data demonstrate the impressive competency of ImDyS in human inverse dynamics and ground reaction force estimation. Moreover, the potential of ImDy(-S) as a fundamental motion analysis tool is exhibited with downstream applications. The project page is https://foruck.github.io/ImDy/.

We study the dynamics in high dimensions of online stochastic gradient descent for the multi-spiked tensor model. This multi-index model arises from the tensor principal component analysis (PCA) problem with multiple spikes, where the goal is to estimate $r$ unknown signal vectors within the $N$-dimensional unit sphere through maximum likelihood estimation from noisy observations of a $p$-tensor. We determine the number of samples and the conditions on the signal-to-noise ratios (SNRs) required to efficiently recover the unknown spikes from natural random initializations. We show that full recovery of all spikes is possible provided a number of sample scaling as $N^{p-2}$, matching the algorithmic threshold identified in the rank-one case [Ben Arous, Gheissari, Jagannath 2020, 2021]. Our results are obtained through a detailed analysis of a low-dimensional system that describes the evolution of the correlations between the estimators and the spikes, while controlling the noise in the dynamics. We find that the spikes are recovered sequentially in a process we term "sequential elimination": once a correlation exceeds a critical threshold, all correlations sharing a row or column index become sufficiently small, allowing the next correlation to grow and become macroscopic. The order in which correlations become macroscopic depends on their initial values and the corresponding SNRs, leading to either exact recovery or recovery of a permutation of the spikes. In the matrix case, when $p=2$, if the SNRs are sufficiently separated, we achieve exact recovery of the spikes, whereas equal SNRs lead to recovery of the subspace spanned by the spikes.

In this paper, we study the problem of low-rank tensor learning, where only a few of training samples are observed and the underlying tensor has a low-rank structure. The existing methods are based on the sum of nuclear norms of unfolding matrices of a tensor, which may be suboptimal. In order to explore the low-rankness of the underlying tensor effectively, we propose a nonconvex model based on transformed tensor nuclear norm for low-rank tensor learning. Specifically, a family of nonconvex functions are employed onto the singular values of all frontal slices of a tensor in the transformed domain to characterize the low-rankness of the underlying tensor. An error bound between the stationary point of the nonconvex model and the underlying tensor is established under restricted strong convexity on the loss function (such as least squares loss and logistic regression) and suitable regularity conditions on the nonconvex penalty function. By reformulating the nonconvex function into the difference of two convex functions, a proximal majorization-minimization (PMM) algorithm is designed to solve the resulting model. Then the global convergence and convergence rate of PMM are established under very mild conditions. Numerical experiments are conducted on tensor completion and binary classification to demonstrate the effectiveness of the proposed method over other state-of-the-art methods.

A murder investigation has been opened after a 16-year-old boy was found fatally injured in north London. Deonte Mowatt-Slater died after his motorbike hit a lamppost following a suspected stabbing. Met Police officers were called in the early hours of Tuesday to reports of a crash on Beachcroft Way in Islington. Medics at the scene said he was found to have a suspected knife injury, and despite attempts to save his life, he died. There have been no arrests and inquiries are ongoing.

Topic models are a popular approach for extracting semantic information from large document collections. However, recent studies suggest that the topics generated by these models often do not align well with human intentions. While metadata such as labels and authorship information is available, it has not yet been effectively incorporated into neural topic models. To address this gap, we introduce FANToM, a novel method for aligning neural topic models with both labels and authorship information. FANToM allows for the inclusion of this metadata when available, producing interpretable topics and author distributions for each topic. Our approach demonstrates greater expressiveness than conventional topic models by learning the alignment between labels, topics, and authors. Experimental results show that FANToM improves upon existing models in terms of both topic quality and alignment. Additionally, it identifies author interests and similarities.

A standard practice when using large language models is for users to supplement their instruction with an input context containing new information for the model to process. However, models struggle to reliably follow the input context, especially when it conflicts with their parametric knowledge from pretraining. In-principle, one would expect models to adapt to the user context better after instruction finetuning, particularly when handling knowledge conflicts. However, we observe a surprising failure mode: during instruction tuning, the context reliance under knowledge conflicts initially increases as expected, but then gradually decreases as instruction finetuning progresses. This happens while the performance on standard benchmarks keeps on increasing far after this drop. We call this phenomenon context-parametric inversion and observe it across multiple general purpose instruction tuning datasets such as TULU, Alpaca and Ultrachat, across different model families like Llama, Mistral, and Pythia. We perform various controlled studies and theoretical analysis to show that context-parametric inversion occurs due to examples in the instruction finetuning data where the input context provides information that aligns with model's parametric knowledge. Our analysis suggests some natural mitigation strategies with limited but insightful gains, and serves as a useful starting point in addressing this deficiency in instruction finetuning.