Jason Brown

Matthew Esmaili Mallory, Mark Glickman, Jason Brown

Statistical modeling of popular music presents a unique challenge due to the complexity of song structures, which cannot be easily analyzed using conventional statistical tools. However, recent advances in data science have shown that converting non-standard data objects into real vector-valued embeddings enables meaningful statistical analysis. In this work, we demonstrate an approach based on logistic principal component analysis to construct embeddings from global song features, allowing for standard multivariate analysis. We apply this method to a corpus of Lennon and McCartney songs from 1962-1966, using embeddings derived from chords, melodic notes, chord and pitch transitions, and melodic contours. Our analysis explores how these song embeddings cluster by Beatles album, how songwriting styles evolved over time, and whether Lennon and McCartney's compositions exhibited convergence or divergence. This embedding-based approach offers a powerful framework for statistically examining musical structure and stylistic development in popular music.

Timothy Pistotti, Jason Brown, Michael Witbrock

Recent studies employing Large Language Models (LLMs) to test the Argument from the Poverty of the Stimulus (APS) have yielded contrasting results across syntactic phenomena. This paper investigates the hypothesis that characteristics of the stimuli used in recent studies, including lexical ambiguities and structural complexities, may confound model performance. A methodology is proposed for re-evaluating LLM competence on syntactic prediction, focusing on GPT-2. This involves: 1) establishing a baseline on previously used (both filtered and unfiltered) stimuli, and 2) generating a new, refined dataset using a state-of-the-art (SOTA) generative LLM (Gemini 2.5 Pro Preview) guided by linguistically-informed templates designed to mitigate identified confounds. Our preliminary findings indicate that GPT-2 demonstrates notably improved performance on these refined PG stimuli compared to baselines, suggesting that stimulus quality significantly influences outcomes in surprisal-based evaluations of LLM syntactic competency.

Timothy Pistotti, Jason Brown, Michael Witbrock

Recent studies probing the Argument from the Poverty of the Stimulus (APS) have applied Large Language Models (LLMs) to test the learnability of complex syntax through surprisal-based metrics. However, divergent conclusions raise questions concerning the insights these metrics offer. While Wilcox et al. (2024) used direct minimal pair comparisons (the "wh-effect") to demonstrate that models successfully generalise knowledge of filler-gap dependencies, Lan et al. (2024) used a Difference-in-Differences (DiD) metric and found that models largely fail on parasitic gaps (PGs). This paper argues that the direct minimal pair approach offers greater diagnostic transparency. We demonstrate this by generating a full 8-permutation paradigm of refined PG stimuli and evaluating the GPT-2 model used in previous studies with a systematic Wilcox-style wh-effect analysis. Our results show that GPT-2 succeeds across all four tested conditions, indicating robust knowledge of filler-gap licensing principles even in complex PG environments. This finding, which contrasts with the more ambiguous results from DiD-style metrics, suggests that the choice of evaluation metric is critical for assessing an LLM's syntactic competence.

Siyu Yu, Zihan Qin, Tingshan Liu et al.

The complete connectome of the Drosophila larva brain offers a unique opportunity to investigate whether biologically evolved circuits can support artificial intelligence. We convert this wiring diagram into a Biological Processing Unit (BPU), a fixed recurrent network derived directly from synaptic connectivity. Despite its modest size 3,000 neurons and 65,000 weights between them), the unmodified BPU achieves 98% accuracy on MNIST and 58% on CIFAR-10, surpassing size-matched MLPs. Scaling the BPU via structured connectome expansions further improves CIFAR-10 performance, while modality-specific ablations reveal the uneven contributions of different sensory subsystems. On the ChessBench dataset, a lightweight GNN-BPU model trained on only 10,000 games achieves 60% move accuracy, nearly 10x better than any size transformer. Moreover, CNN-BPU models with ~2M parameters outperform parameter-matched Transformers, and with a depth-6 minimax search at inference, reach 91.7% accuracy, exceeding even a 9M-parameter Transformer baseline. These results demonstrate the potential of biofidelic neural architectures to support complex cognitive tasks and motivate scaling to larger and more intelligent connectomes in future work.

Jason Brown, Bohan Chen, Harris Hardiman-Mostow et al.

Standard deep learning architectures used for classification generate label predictions with a projection head and softmax activation function. Although successful, these methods fail to leverage the relational information between samples in the batch for generating label predictions. In recent works, graph-based learning techniques, namely Laplace learning, have been heuristically combined with neural networks for both supervised and semi-supervised learning (SSL) tasks. However, prior works approximate the gradient of the loss function with respect to the graph learning algorithm or decouple the processes; end-to-end integration with neural networks is not achieved. In this work, we derive backpropagation equations, via the adjoint method, for inclusion of a general family of graph learning layers into a neural network. This allows us to precisely integrate graph Laplacian-based label propagation into a neural network layer, replacing a projection head and softmax activation function for classification tasks. Using this new framework, our experimental results demonstrate smooth label transitions across data, improved robustness to adversarial attacks, improved generalization, and improved training dynamics compared to the standard softmax-based approach.

Jacob Shermeyer, Daniel Hogan, Jason Brown et al.

Within the remote sensing domain, a diverse set of acquisition modalities exist, each with their own unique strengths and weaknesses. Yet, most of the current literature and open datasets only deal with electro-optical (optical) data for different detection and segmentation tasks at high spatial resolutions. optical data is often the preferred choice for geospatial applications, but requires clear skies and little cloud cover to work well. Conversely, Synthetic Aperture Radar (SAR) sensors have the unique capability to penetrate clouds and collect during all weather, day and night conditions. Consequently, SAR data are particularly valuable in the quest to aid disaster response, when weather and cloud cover can obstruct traditional optical sensors. Despite all of these advantages, there is little open data available to researchers to explore the effectiveness of SAR for such applications, particularly at very-high spatial resolutions, i.e. <1m Ground Sample Distance (GSD). To address this problem, we present an open Multi-Sensor All Weather Mapping (MSAW) dataset and challenge, which features two collection modalities (both SAR and optical). The dataset and challenge focus on mapping and building footprint extraction using a combination of these data sources. MSAW covers 120 km^2 over multiple overlapping collects and is annotated with over 48,000 unique building footprints labels, enabling the creation and evaluation of mapping algorithms for multi-modal data. We present a baseline and benchmark for building footprint extraction with SAR data and find that state-of-the-art segmentation models pre-trained on optical data, and then trained on SAR (F1 score of 0.21) outperform those trained on SAR data alone (F1 score of 0.135).