Lying at the interface between Network Science and Machine Learning, node embedding algorithms take a graph as input and encode its structure onto output vectors that represent nodes in an abstract geometric space, enabling various vector-based downstream tasks such as network modelling, data compression, link prediction, and community detection. Two apparently unrelated limitations affect these algorithms. On one hand, it is not clear what the basic operation defining vector spaces, i.e. the vector sum, corresponds to in terms of the original nodes in the network. On the other hand, while the same input network can be represented at multiple levels of resolution by coarse-graining the constituent nodes into arbitrary block-nodes, the relationship between node embeddings obtained at different hierarchical levels is not understood. Here, building on recent results in network renormalization theory, we address these two limitations at once and define a multiscale node embedding method that, upon arbitrary coarse-grainings, ensures statistical consistency of the embedding vector of a block-node with the sum of the embedding vectors of its constituent nodes. We illustrate the power of this approach on two economic networks that can be naturally represented at multiple resolution levels: namely, the international trade between (sets of) countries and the input-output flows among (sets of) industries in the Netherlands. We confirm the statistical consistency between networks retrieved from coarse-grained node vectors and networks retrieved from sums of fine-grained node vectors, a result that cannot be achieved by alternative methods. Several key network properties, including a large number of triangles, are successfully replicated already from embeddings of very low dimensionality, allowing for the generation of faithful replicas of the original networks at arbitrary resolution levels.

This paper introduces the counter-intuitive generalization results of overfitting pre-trained large language models (LLMs) on very small datasets. In the setting of open-ended text generation, it is well-documented that LLMs tend to generate repetitive and dull sequences, a phenomenon that is especially apparent when generating using greedy decoding. This issue persists even with state-of-the-art LLMs containing billions of parameters, trained via next-token prediction on large datasets. We find that by further fine-tuning these models to achieve a near-zero training loss on a small set of samples - a process we refer to as hyperfitting - the long-sequence generative capabilities are greatly enhanced. Greedy decoding with these Hyperfitted models even outperform Top-P sampling over long-sequences, both in terms of diversity and human preferences. This phenomenon extends to LLMs of various sizes, different domains, and even autoregressive image generation. We further find this phenomena to be distinctly different from that of Grokking and double descent. Surprisingly, our experiments indicate that hyperfitted models rarely fall into repeating sequences they were trained on, and even explicitly blocking these sequences results in high-quality output. All hyperfitted models produce extremely low-entropy predictions, often allocating nearly all probability to a single token. Despite the recent rapid advancements in artificial intelligence spearheaded by Transformer-based large language models (LLMs) and their emergent phenomena (Wei et al., 2022b; Bubeck et al., 2023), models trained on next-token pre-training objectives often degenerate when producing longer texts. This is particularly true for greedy decoding, and has resulted in mitigation strategies such as repetition penalties (Keskar et al., 2019) and nucleus sampling (Holtzman et al., 2020). However, when removing these heuristics and simply picking the top-1 candidate at each time-step, LLMs display strong tendencies to repeat themselves at the token, phrase, and sentence level (Holtzman et al., 2020), as is exemplified in Figure 1. This is a recurrent phenomenon for which there are many proposed hypotheses but, to the best of our knowledge, no definitive explanation exists. Color indicating how repetitive the generated text is. Although these models achieve significantly worse validation loss, they produce texts that align markedly better with human preferences and automatic diversity metrics. Indeed, we find that hyperfitting state-of-the-art LLMs yields capabilities that outperform models with 10x the number of parameters.

The application of deep learning methods, particularly foundation models, in biological research has surged in recent years. These models can be text-based or trained on underlying biological data, especially omics data of various types. However, comparing the performance of these models consistently has proven to be a challenge due to differences in training data and downstream tasks. To tackle this problem, we developed an architecture-agnostic benchmarking approach that, instead of evaluating the models directly, leverages entity representation vectors from each model and trains simple predictive models for each benchmarking task. This ensures that all types of models are evaluated using the same input and output types. Here we focus on gene properties collected from professionally curated bioinformatics databases. These gene properties are categorized into five major groups: genomic properties, regulatory functions, localization, biological processes, and protein properties. Overall, we define hundreds of tasks based on these databases, which include binary, multi-label, and multi-class classification tasks. We apply these benchmark tasks to evaluate expression-based models, large language models, protein language models, DNA-based models, and traditional baselines. Our findings suggest that text-based models and protein language models generally outperform expression-based models in genomic properties and regulatory functions tasks, whereas expression-based models demonstrate superior performance in localization tasks. These results should aid in the development of more informed artificial intelligence strategies for biological understanding and therapeutic discovery. To ensure the reproducibility and transparency of our findings, we have made the source code and benchmark data publicly accessible for further investigation and expansion at github.com/BiomedSciAI/gene-benchmark.

Numerous politicians use social media platforms, particularly X, to engage with their constituents. This interaction allows constituents to pose questions and offer feedback but also exposes politicians to a barrage of hostile responses, especially given the anonymity afforded by social media. They are typically targeted in relation to their governmental role, but the comments also tend to attack their personal identity. This can discredit politicians and reduce public trust in the government. It can also incite anger and disrespect, leading to offline harm and violence. While numerous models exist for detecting hostility in general, they lack the specificity required for political contexts. Furthermore, addressing hostility towards politicians demands tailored approaches due to the distinct language and issues inherent to each country (e.g., Brexit for the UK). To bridge this gap, we construct a dataset of 3,320 English tweets spanning a two-year period manually annotated for hostility towards UK MPs. Our dataset also captures the targeted identity characteristics (race, gender, religion, none) in hostile tweets. We perform linguistic and topical analyses to delve into the unique content of the UK political data. Finally, we evaluate the performance of pre-trained language models and large language models on binary hostility detection and multi-class targeted identity type classification tasks. Our study offers valuable data and insights for future research on the prevalence and nature of politics-related hostility specific to the UK.

The rapid advancement of artificial intelligence (AI) technologies presents profound challenges to societal safety. As AI systems become more capable, accessible, and integrated into critical services, the dual nature of their potential is increasingly clear. While AI can enhance defensive capabilities in areas like threat detection, risk assessment, and automated security operations (Hassanin and Moustafa, 2024), it also presents avenues for malicious exploitation and largescale societal harm, for example through automated influence operations and cyber attacks (Goldstein et al., 2023; Xu et al., 2024a). Understanding the dynamics that shape AI's capacity to both cause harm and enhance protective measures is essential for informed decision-making regarding the deployment, use, and integration of advanced AI systems. This paper builds on recent work on offense-defense dynamics within the realm of AI (Schneier, 2018; Garfinkel and Dafoe, 2021), proposing a taxonomy to map and examine the key factors that influence whether AI systems predominantly pose threats or offer protective benefits to society. By establishing a shared terminology and conceptual foundation for analyzing these interactions, this work seeks to facilitate further research and discourse in this critical area.

Multi-task large language models (MTLLMs) are important for many applications at the wireless edge, where users demand specialized models to handle multiple tasks efficiently. However, training MTLLMs is complex and exhaustive, particularly when tasks are subject to change. Recently, the concept of model fusion via task vectors has emerged as an efficient approach for combining fine-tuning parameters to produce an MTLLM. In this paper, the problem of enabling edge users to collaboratively craft such MTLMs via tasks vectors is studied, under the assumption of worst-case adversarial attacks. To this end, first the influence of adversarial noise to multi-task model fusion is investigated and a relationship between the so-called weight disentanglement error and the mean squared error (MSE) is derived. Using hypothesis testing, it is directly shown that the MSE increases interference between task vectors, thereby rendering model fusion ineffective. Then, a novel resilient MTLLM fusion (R-MTLLMF) is proposed, which leverages insights about the LLM architecture and fine-tuning process to safeguard task vector aggregation under adversarial noise by realigning the MTLLM. The proposed R-MTLLMF is then compared for both worst-case and ideal transmission scenarios to study the impact of the wireless channel. Extensive model fusion experiments with vision LLMs demonstrate R-MTLLMF's effectiveness, achieving close-to-baseline performance across eight different tasks in ideal noise scenarios and significantly outperforming unprotected model fusion in worst-case scenarios. The results further advocate for additional physical layer protection for a holistic approach to resilience, from both a wireless and LLM perspective.

Large Language Models (LLMs) have demonstrated remarkable capabilities in generating human-like text, but their output may not be aligned with the user or even produce harmful content. This paper presents a novel approach to detect and steer concepts such as toxicity before generation. We introduce the Sparse Conditioned Autoencoder (SCAR), a single trained module that extends the otherwise untouched LLM. SCAR ensures full steerability, towards and away from concepts (e.g., toxic content), without compromising the quality of the model's text generation on standard evaluation benchmarks. We demonstrate the effective application of our approach through a variety of concepts, including toxicity, safety, and writing style alignment. As such, this work establishes a robust framework for controlling LLM generations, ensuring their ethical and safe deployment in real-world applications.

Safety alignment of Large Language Models (LLMs) has recently become a critical objective of model developers. In response, a growing body of work has been investigating how safety alignment can be bypassed through various jailbreaking methods, such as adversarial attacks. However, these jailbreak methods can be rather costly or involve a non-trivial amount of creativity and effort, introducing the assumption that malicious users are high-resource or sophisticated. In this paper, we study how simple random augmentations to the input prompt affect safety alignment effectiveness in state-of-the-art LLMs, such as Llama 3 and Qwen 2. We perform an in-depth evaluation of 17 different models and investigate the intersection of safety under random augmentations with multiple dimensions: augmentation type, model size, quantization, fine-tuning-based defenses, and decoding strategies (e.g., sampling temperature). We show that low-resource and unsophisticated attackers, i.e. $\textit{stochastic monkeys}$, can significantly improve their chances of bypassing alignment with just 25 random augmentations per prompt. Source code and data: https://github.com/uiuc-focal-lab/stochastic-monkeys/

Recent efforts in fine-tuning language models often rely on automatic data selection, commonly using Nearest Neighbors retrieval from large datasets. However, we theoretically show that this approach tends to select redundant data, limiting its effectiveness or even hurting performance. To address this, we introduce SIFT, a data selection algorithm designed to reduce uncertainty about the model's response given a prompt, which unifies ideas from retrieval and active learning. Whereas Nearest Neighbor retrieval typically fails in the presence of information duplication, SIFT accounts for information duplication and optimizes the overall information gain of the selected examples. We focus our evaluations on fine-tuning at test-time for prompt-specific language modeling on the Pile dataset, and show that SIFT consistently outperforms Nearest Neighbor retrieval, with minimal computational overhead. Moreover, we show that our uncertainty estimates can predict the performance gain of test-time fine-tuning, and use this to develop an adaptive algorithm that invests test-time compute proportional to realized performance gains. We provide the $\texttt{activeft}$ (Active Fine-Tuning) library which can be used as a drop-in replacement for Nearest Neighbor retrieval.

Thailand has targeted to achieve carbon neutrality by 2050 when the power grid will need to accommodate 50% share of renewable electricity generation capacity; see [Ene21]. The most recent draft of Power Development Plan 2024 (PDP2024) for 2024 - 2037 from [Ene24] proposes to add a new solar generation capacity of approximately 24,400 MWp (more than 4 times the amount issued in the previous Alternative Energy Development Plan 2015-2036 (AEDP2015) at 6,000 MWp, shown in [Dep15, p.9]. This amount does not yet include behind-the-meter, self-generation solar installed capacities of the prosumers, which is expected to increase at an accelerating rate. Solar integration into the power grid with such a sharprising amount will pose technical challenges to the operation and control of the transmission and distribution networks, carried out by the transmission system operator (TSO) and distribution system operator (DSO), as presented in [OB16]. Hence, TSO in Thailand will need an effective means to estimate the solar power generation across the entire transmission network, on an hourly basis, or even finer time resolution, to provide economic hour-to-hour generation dispatch for load following the total net load of the transmission, and to prepare sufficient system flexibility (i.e., ramp-rate capability of the thermal and hydropower plants, or energy storage systems) to cope with the net load fluctuation due to solar generation intermittency for maintaining system frequency stability, concurrently, in its operation. For DSO, a significant amount of reverse power flow when self-generation from solar exceeds self-consumption can lead to technical concerns of voltage regulation and equipment overloading problems. The near real-time estimation of solar generation in each distribution area will enable DSO to activate proper network switching or reconfiguring to mitigate such fundamental concerns to ensure its reliable operation.