Modeling voice identity is challenging due to its multifaceted nature. In generative speech systems, identity is often assessed using automatic speaker verification (ASV) embeddings, designed for discrimination rather than characterizing identity. This paper investigates which aspects of a voice are captured in such representations. We find that widely used ASV embeddings focus mainly on static features like timbre and pitch range, while neglecting dynamic elements such as rhythm. We also identify confounding factors that compromise speaker similarity measurements and suggest mitigation strategies. To address these gaps, we propose U3D, a metric that evaluates speakers' dynamic rhythm patterns. This work contributes to the ongoing challenge of assessing speaker identity consistency in the context of ever-better voice cloning systems. We publicly release our code.

Large Language Models (LLMs) are increasingly being integrated into various applications and services running on billions of mobile devices. However, deploying LLMs on resource-limited mobile devices faces a significant challenge due to their high demand for computation, memory, and ultimately energy. While current LLM frameworks for mobile use three power-hungry components-CPU, GPU, and Memory-even when running primarily-GPU LLM models, optimized DVFS governors for CPU, GPU, and memory featured in modern mobile devices operate independently and are oblivious of each other. Motivated by the above observation, in this work, we first measure the energy-efficiency of a SOTA LLM framework consisting of various LLM models on mobile phones which showed the triplet mobile governors result in up to 40.4% longer prefilling and decoding latency compared to optimal combinations of CPU, GPU, and memory frequencies with the same energy consumption for sampled prefill and decode lengths. Second, we conduct an in-depth measurement study to uncover how the intricate interplay (or lack of) among the mobile governors cause the above inefficiency in LLM inference. Finally, based on these insights, we design FUSE - a unified energy-aware governor for optimizing the energy efficiency of LLM inference on mobile devices. Our evaluation using a ShareGPT dataset shows FUSE reduces the time-to-first-token and time-per-output-token latencies by 7.0%-16.9% and 25.4%-36.8% on average with the same energy-per-token for various mobile LLM models.

Cross-subject motor imagery (CS-MI) classification in brain-computer interfaces (BCIs) is a challenging task due to the significant variability in Electroencephalography (EEG) patterns across different individuals. This variability often results in lower classification accuracy compared to subject-specific models, presenting a major barrier to developing calibration-free BCIs suitable for real-world applications. In this paper, we introduce a novel approach that significantly enhances cross-subject MI classification performance through optimized preprocessing and deep learning techniques. Our approach involves direct classification of Short-Time Fourier Transform (STFT)-transformed EEG data, optimized STFT parameters, and a balanced batching strategy during training of a Convolutional Neural Network (CNN). This approach is uniquely validated across four different datasets, including three widely-used benchmark datasets leading to substantial improvements in cross-subject classification, achieving 67.60% on the BCI Competition IV Dataset 1 (IV-1), 65.96% on Dataset 2A (IV-2A), and 80.22% on Dataset 2B (IV-2B), outperforming state-of-the-art techniques. Additionally, we systematically investigate the classification performance using MI windows ranging from the full 4-second window to 1-second windows. These results establish a new benchmark for generalizable, calibration-free MI classification in addition to contributing a robust open-access dataset to advance research in this domain.

Existing visual model-based reinforcement learning (MBRL) algorithms with observation reconstruction often suffer from information conflicts, making it difficult to learn compact representations and hence result in less robust policies, especially in the presence of task-irrelevant visual distractions. In this paper, we first reveal that the information conflicts in current visual MBRL algorithms stem from visual representation learning and latent dynamics modeling with an information-theoretic perspective. Based on this finding, we present a new algorithm to resolve information conflicts for visual MBRL, named MInCo, which mitigates information conflicts by leveraging negative-free contrastive learning, aiding in learning invariant representation and robust policies despite noisy observations. To prevent the dominance of visual representation learning, we introduce time-varying reweighting to bias the learning towards dynamics modeling as training proceeds. We evaluate our method on several robotic control tasks with dynamic background distractions. Our experiments demonstrate that MInCo learns invariant representations against background noise and consistently outperforms current state-of-the-art visual MBRL methods. Code is available at https://github.com/ShiguangSun/minco.

Layer-wise normalization (LN) is an essential component of virtually all transformer-based large language models. While its effects on training stability are well documented, its role at inference time is poorly understood. Additionally, LN layers hinder mechanistic interpretability by introducing additional nonlinearities and increasing the interconnectedness of individual model components. Here, we show that all LN layers can be removed from every GPT-2 model with only a small increase in validation loss (e.g. +0.03 cross-entropy loss for GPT-2 XL). Thus, LN cannot play a substantial role in language modeling. We find that the amount of fine-tuning data needed for LN removal grows sublinearly with model parameters, suggesting scaling to larger models is feasible. We release a suite of LN-free GPT-2 models on Hugging Face. Furthermore, we test interpretability techniques on LN-free models. Direct logit attribution now gives the exact direct effect of individual components, while the accuracy of attribution patching does not significantly improve. We also confirm that GPT-2's "confidence neurons" are inactive in the LN-free models. Our work clarifies the role of LN layers in language modeling, showing that GPT-2-class models can function without LN layers. We hope that our LN-free analogs of the GPT-2 family of models will enable more precise interpretability research and improve our understanding of language models.

This paper describes our system for the SciVQA 2025 Shared Task on Scientific Visual Question Answering. Our system employs an ensemble of two Multimodal Large Language Models and various few-shot example retrieval strategies. The model and few-shot setting are selected based on the figure and question type. We also select answers based on the models' confidence levels. On the blind test data, our system ranks third out of seven with an average F1 score of 85.12 across ROUGE-1, ROUGE-L, and BERTS. Our code is publicly available.

We present a theoretical framework for deriving the general $n$-th order Fréchet derivatives of singular values in real rectangular matrices, by leveraging reduced resolvent operators from Kato's analytic perturbation theory for self-adjoint operators. Deriving closed-form expressions for higher-order derivatives of singular values is notoriously challenging through standard matrix-analysis techniques. To overcome this, we treat a real rectangular matrix as a compact operator on a finite-dimensional Hilbert space, and embed the rectangular matrix into a block self-adjoint operator so that non-symmetric perturbations are captured. Applying Kato's asymptotic eigenvalue expansion to this construction, we obtain a general, closed-form expression for the infinitesimal $n$-th order spectral variations. Specializing to $n=2$ and deploying on a Kronecker-product representation with matrix convention yield the Hessian of a singular value, not found in literature. By bridging abstract operator-theoretic perturbation theory with matrices, our framework equips researchers with a practical toolkit for higher-order spectral sensitivity studies in random matrix applications (e.g., adversarial perturbation in deep learning).

The field of hypothesis generation promises to reduce costs in neuroscience by narrowing the range of interventional studies needed to study various phenomena. Existing machine learning methods can generate scientific hypotheses from complex datasets, but many approaches assume causal relationships are static over time, limiting their applicability to systems with dynamic, state-dependent behavior, such as the brain. While some techniques attempt dynamic causal discovery through factor models, they often restrict relationships to linear patterns or impose other simplifying assumptions. We propose a novel method that models dynamic graphs as a conditionally weighted superposition of static graphs, where each static graph can capture nonlinear relationships. This approach enables the detection of complex, time-varying interactions between variables beyond linear limitations. Our method improves f1-scores of predicted dynamic causal patterns by roughly 22-28% on average over baselines in some of our experiments, with some improvements reaching well over 60%. A case study on real brain data demonstrates our method's ability to uncover relationships linked to specific behavioral states, offering valuable insights into neural dynamics.

Investors and stock market analysts face major challenges in predicting stock returns and making wise investment decisions. The predictability of equity stock returns can boost investor confidence, but it remains a difficult task. To address this issue, a study was conducted using a Long Short-term Memory (LSTM) model to predict future stock market movements. The study used a historical dataset from the Nigerian Stock Exchange (NSE), which was cleaned and normalized to design the LSTM model. The model was evaluated using performance metrics and compared with other deep learning models like Artificial and Convolutional Neural Networks (CNN). The experimental results showed that the LSTM model can predict future stock market prices and returns with over 90% accuracy when trained with a reliable dataset. The study concludes that LSTM models can be useful in predicting financial time-series-related problems if well-trained. Future studies should explore combining LSTM models with other deep learning techniques like CNN to create hybrid models that mitigate the risks associated with relying on a single model for future equity stock predictions.

W e present a novel and detailed dataset on origin-destination annual migration flows and stocks between 230 countries and regions, spanning the period from 1990 to the present. Our flow estimates are further disaggregated by country of birth, providing a comprehensive picture of migration over the last 35 years. The estimates are obtained by training a deep recurrent neural network to learn flow patterns from 18 covariates for all countries, including geographic, economic, cultural, societal, and political information. The recurrent architecture of the neural network means that the entire past can influence current migration patterns, allowing us to learn long-range temporal correlations. By training an ensemble of neural networks and additionally pushing uncertainty on the covariates through the trained network, we obtain confidence bounds for all our estimates, allowing researchers to pinpoint the geographic regions most in need of additional data collection. W e validate our approach on various test sets of unseen data, demonstrating that it significantly outperforms traditional methods estimating five-year flows while delivering a significant increase in temporal resolution. The model is fully open source: all training data, neural network weights, and training code are made public alongside the migration estimates, providing a valuable resource for future studies of human migration.