The strict selectivity of the blood-brain barrier (BBB) represents one of the most formidable challenges to successful central nervous system (CNS) drug delivery. Computational methods to generate BBB permeable drugs in silico may be valuable tools in the CNS drug design pipeline. However, in real-world applications, BBB penetration alone is insufficient; rather, after transiting the BBB, molecules must bind to a specific target or receptor in the brain and must also be safe and non-toxic. To discover small molecules that concurrently satisfy these constraints, we use multi-objective generative AI to synthesize drug-like BBB-permeable small molecules. Specifically, we computationally synthesize molecules with predicted binding affinity against dopamine receptor D2, the primary target for many clinically effective antipsychotic drugs. After training several graph neural network-based property predictors, we adapt SyntheMol (Swanson et al., 2024), a recently developed Monte Carlo Tree Search-based algorithm for antibiotic design, to perform a multi-objective guided traversal over an easily synthesizable molecular space. We design a library of 26,581 novel and diverse small molecules containing hits with high predicted BBB permeability and favorable predicted safety and toxicity profiles, and that could readily be synthesized for experimental validation in the wet lab. We also validate top scoring molecules with molecular docking simulation against the D2 receptor and demonstrate predicted binding affinity on par with risperidone, a clinically prescribed D2-targeting antipsychotic. In the future, the SyntheMol-based computational approach described here may enable the discovery of novel neurotherapeutics for currently intractable disorders of the CNS.

Large language models (LLMs) have shown remarkable performance in various natural language processing tasks. However, a primary constraint they face is the context limit, i.e., the maximum number of tokens they can process. Previous works have explored architectural changes and modifications in positional encoding to relax the constraint, but they often require expensive training or do not address the computational demands of self-attention. In this paper, we present Hierarchical cOntext MERging (HOMER), a new training-free scheme designed to overcome the limitations. HOMER uses a divide-and-conquer algorithm, dividing long inputs into manageable chunks. Each chunk is then processed collectively, employing a hierarchical strategy that merges adjacent chunks at progressive transformer layers. A token reduction technique precedes each merging, ensuring memory usage efficiency. We also propose an optimized computational order reducing the memory requirement to logarithmically scale with respect to input length, making it especially favorable for environments with tight memory restrictions. Our experiments demonstrate the proposed method's superior performance and memory efficiency, enabling the broader use of LLMs in contexts requiring extended context. Code is available at https://github.com/alinlab/HOMER.

Despite being trained specifically to follow user instructions, today's language models perform poorly when instructed to produce random outputs. For example, when prompted to pick a number uniformly between one and ten Llama-2-13B-chat disproportionately favors the number five, and when tasked with picking a first name at random, Mistral-7B-Instruct chooses Avery 40 times more often than we would expect based on the U.S. population. When these language models are used for real-world tasks where diversity of outputs is crucial, such as language model assisted dataset construction, their inability to produce diffuse distributions over valid choices is a major hurdle. In this work, we propose a fine-tuning method that encourages language models to output distributions that are diffuse over valid outcomes. The methods we introduce generalize across a variety of tasks and distributions and make large language models practical for synthetic dataset generation with little human intervention.

As large language models (LLMs) become ubiquitous in our daily tasks and digital interactions, associated privacy risks are increasingly in focus. While LLM privacy research has primarily focused on the leakage of model training data, it has recently been shown that the increase in models' capabilities has enabled LLMs to make accurate privacy-infringing inferences from previously unseen texts. With the rise of multimodal vision-language models (VLMs), capable of understanding both images and text, a pertinent question is whether such results transfer to the previously unexplored domain of benign images posted online. To investigate the risks associated with the image reasoning capabilities of newly emerging VLMs, we compile an image dataset with human-annotated labels of the image owner's personal attributes. In order to understand the additional privacy risk posed by VLMs beyond traditional human attribute recognition, our dataset consists of images where the inferable private attributes do not stem from direct depictions of humans. On this dataset, we evaluate the inferential capabilities of 7 state-of-the-art VLMs, finding that they can infer various personal attributes at up to 77.6% accuracy. Concerningly, we observe that accuracy scales with the general capabilities of the models, implying that future models can be misused as stronger adversaries, establishing an imperative for the development of adequate defenses.

Socioeconomic bias in society exacerbates disparities, influencing access to opportunities and resources based on individuals' economic and social backgrounds. This pervasive issue perpetuates systemic inequalities, hindering the pursuit of inclusive progress as a society. In this paper, we investigate the presence of socioeconomic bias, if any, in large language models. To this end, we introduce a novel dataset SilverSpoon, consisting of 3000 samples that illustrate hypothetical scenarios that involve underprivileged people performing ethically ambiguous actions due to their circumstances, and ask whether the action is ethically justified. Further, this dataset has a dual-labeling scheme and has been annotated by people belonging to both ends of the socioeconomic spectrum. Using SilverSpoon, we evaluate the degree of socioeconomic bias expressed in large language models and the variation of this degree as a function of model size. We also perform qualitative analysis to analyze the nature of this bias. Our analysis reveals that while humans disagree on which situations require empathy toward the underprivileged, most large language models are unable to empathize with the socioeconomically underprivileged regardless of the situation. To foster further research in this domain, we make SilverSpoon and our evaluation harness publicly available.

This paper describes AssemblyAI's industrial-scale automatic speech recognition (ASR) system, designed to meet the requirements of large-scale, multilingual ASR serving various application needs. Our system leverages a diverse training dataset comprising unsupervised (12.5M hours), supervised (188k hours), and pseudo-labeled (1.6M hours) data across four languages. We provide a detailed description of our model architecture, consisting of a full-context 600M-parameter Conformer encoder pre-trained with BEST-RQ and an RNN-T decoder fine-tuned jointly with the encoder. Our extensive evaluation demonstrates competitive word error rates (WERs) against larger and more computationally expensive models, such as Whisper large and Canary-1B. Furthermore, our architectural choices yield several key advantages, including an improved code-switching capability, a 5x inference speedup compared to an optimized Whisper baseline, a 30% reduction in hallucination rate on speech data, and a 90% reduction in ambient noise compared to Whisper, along with significantly improved time-stamp accuracy. Throughout this work, we adopt a system-centric approach to analyzing various aspects of fully-fledged ASR models to gain practically relevant insights useful for real-world services operating at scale.

During their operation, due to shifts in environmental conditions, devices undergo various forms of detuning from their optimal settings. Typically, this is addressed through control loops, which monitor variables and the device performance, to maintain settings at their optimal values. Quantum devices are particularly challenging since their functionality relies on precisely tuning their parameters. At the same time, the detailed modeling of the environmental behavior is often computationally unaffordable, while a direct measure of the parameters defining the system state is costly and introduces extra noise in the mechanism. In this study, we investigate the application of reinforcement learning techniques to develop a model-free control loop for continuous recalibration of quantum device parameters. Furthermore, we explore the advantages of incorporating minimal environmental noise models. As an example, the application to numerical simulations of a Kennedy receiver-based long-distance quantum communication protocol is presented.

In the context of increasing demands for long-term multi-energy load forecasting in real-world applications, this paper introduces Patchformer, a novel model that integrates patch embedding with encoder-decoder Transformer-based architectures. To address the limitation in existing Transformer-based models, which struggle with intricate temporal patterns in long-term forecasting, Patchformer employs patch embedding, which predicts multivariate time-series data by separating it into multiple univariate data and segmenting each of them into multiple patches. This method effectively enhances the model's ability to capture local and global semantic dependencies. The numerical analysis shows that the Patchformer obtains overall better prediction accuracy in both multivariate and univariate long-term forecasting on the novel Multi-Energy dataset and other benchmark datasets. In addition, the positive effect of the interdependence among energy-related products on the performance of long-term time-series forecasting across Patchformer and other compared models is discovered, and the superiority of the Patchformer against other models is also demonstrated, which presents a significant advancement in handling the interdependence and complexities of long-term multi-energy forecasting. Lastly, Patchformer is illustrated as the only model that follows the positive correlation between model performance and the length of the past sequence, which states its ability to capture long-range past local semantic information.

Predicting the future is of great interest across many aspects of human activity. Businesses are interested in future trends, traders are interested in future stock prices, and companies are highly interested in future technological breakthroughs. While there are many automated systems for predicting future numerical data, such as weather, stock prices, and demand for products, there is relatively little work in automatically predicting textual data. Humans are interested in textual data predictions because it is a natural format for our consumption, and experts routinely make predictions in a textual format (Christensen et al., 2004; Tetlock & Gardner, 2015; Frick, 2015). However, there has been relatively little formalization of this general problem in the machine learning or natural language processing communities. To address this gap, we introduce the task of future language modeling: probabilistic modeling of texts in the future based on a temporal history of texts. To our knowledge, our work is the first work to formalize the task of predicting the future in this way. We show that it is indeed possible to build future language models that improve upon strong non-temporal language model baselines, opening the door to working on this important, and widely applicable problem.

Recently, prototype learning has emerged in semi-supervised medical image segmentation and achieved remarkable performance. However, the scarcity of labeled data limits the expressiveness of prototypes in previous methods, potentially hindering the complete representation of prototypes for class embedding. To address this problem, we propose the Mixed Prototype Consistency Learning (MPCL) framework, which includes a Mean Teacher and an auxiliary network. The Mean Teacher generates prototypes for labeled and unlabeled data, while the auxiliary network produces additional prototypes for mixed data processed by CutMix. Through prototype fusion, mixed prototypes provide extra semantic information to both labeled and unlabeled prototypes. High-quality global prototypes for each class are formed by fusing two enhanced prototypes, optimizing the distribution of hidden embeddings used in consistency learning. Extensive experiments on the left atrium and type B aortic dissection datasets demonstrate MPCL's superiority over previous state-of-the-art approaches, confirming the effectiveness of our framework. The code will be released soon.