Singular learning theory (SLT) \citep{watanabe2009algebraic,watanabe2018mathematical} provides a rigorous asymptotic framework for Bayesian models with non-identifiable parameterizations, yet the statistical meaning of its second-order invariant, the \emph{singular fluctuation}, has remained unclear. In this work, we show that singular fluctuation admits a precise and natural interpretation as a \emph{specific heat}: the second derivative of the Bayesian free energy with respect to temperature. Equivalently, it measures the posterior variance of the log-likelihood observable under the tempered Gibbs posterior. We further introduce a collection of related thermodynamic quantities, including entropy flow, prior susceptibility, and cross-susceptibility, that together provide a detailed geometric diagnosis of singular posterior structure. Through extensive numerical experiments spanning discrete symmetries, boundary singularities, continuous gauge freedoms, and piecewise (ReLU) models, we demonstrate that these thermodynamic signatures cleanly distinguish singularity types, exhibit stable finite-sample behavior, and reveal phase-transition--like phenomena as temperature varies. We also show empirically that the widely used WAIC estimator \citep{watanabe2010asymptotic, watanabe2013widely} is exactly twice the thermodynamic specific heat at unit temperature, clarifying its robustness in singular models.Our results establish a concrete bridge between singular learning theory and statistical mechanics, providing both theoretical insight and practical diagnostics for modern Bayesian models.

Classical statistical inference and learning theory often fail to explain the success of modern neural networks. A key reason is that these models are non-identifiable (singular), violating core assumptions behind PAC bounds and asymptotic normality. Singular learning theory (SLT), a physics-inspired framework grounded in algebraic geometry, has gained popularity for its ability to close this theory-practice gap. In this paper, we empirically study SLT in toy settings relevant to interpretability and phase transitions. First, we understand the SLT free energy $\mathcal{F}_n$ by testing an Arrhenius-style rate hypothesis using both a grokking modulo-arithmetic model and Anthropic's Toy Models of Superposition. Second, we understand the local learning coefficient $λ_α$ by measuring how it scales with problem difficulty across several controlled network families (polynomial regressors, low-rank linear networks, and low-rank autoencoders). Our experiments recover known scaling laws while others yield meaningful deviations from theoretical expectations. Overall, our paper illustrates the many merits of SLT for understanding neural network phase transitions, and poses open research questions for the field.

The CHiME-7 and 8 distant speech recognition (DASR) challenges focus on multi-channel, generalizable, joint automatic speech recognition (ASR) and diarization of conversational speech. With participation from 9 teams submitting 32 diverse systems, these challenges have contributed to state-of-the-art research in the field. This paper outlines the challenges' design, evaluation metrics, datasets, and baseline systems while analyzing key trends from participant submissions. From this analysis it emerges that: 1) Most participants use end-to-end (e2e) ASR systems, whereas hybrid systems were prevalent in previous CHiME challenges. This transition is mainly due to the availability of robust large-scale pre-trained models, which lowers the data burden for e2e-ASR. 2) Despite recent advances in neural speech separation and enhancement (SSE), all teams still heavily rely on guided source separation, suggesting that current neural SSE techniques are still unable to reliably deal with complex scenarios and different recording setups. 3) All best systems employ diarization refinement via target-speaker diarization techniques. Accurate speaker counting in the first diarization pass is thus crucial to avoid compounding errors and CHiME-8 DASR participants especially focused on this part. 4) Downstream evaluation via meeting summarization can correlate weakly with transcription quality due to the remarkable effectiveness of large-language models in handling errors. On the NOTSOFAR-1 scenario, even systems with over 50% time-constrained minimum permutation WER can perform roughly on par with the most effective ones (around 11%). 5) Despite recent progress, accurately transcribing spontaneous speech in challenging acoustic environments remains difficult, even when using computationally intensive system ensembles.

Japan's Kioxia sees AI driving strong memory growth for years Kioxia has been aggressively investing in its main chip factories as it aims to close the gap to rivals Samsung and SK Hynix. Kioxia anticipates demand for NAND storage will grow by roughly 20% each year as AI data center operators keep scaling up. The Tokyo-based memory maker is confident that the market will sustain that rapid clip of expansion and is making investment decisions on a monthly basis to ensure its new plant is up to the task of filling the demand, Executive Vice President Tomoharu Watanabe said. Kioxia's specialty is in NAND flash memory, which is used everywhere from smartphones and laptops to the fast-access sections of data center operations. Demand is strong, especially from hyperscalers who need chips for generative AI purposes," Watanabe said on Tuesday. We're also hearing from customers who need to replace data center servers they installed five to six years ago, as well as some saying they can't get enough hard drives."

Tree-structured Parzen estimator (TPE) is a versatile hyperparameter optimization (HPO) method supported by popular HPO tools. Since these HPO tools have been developed in line with the trend of deep learning (DL), the problem setups often used in the DL domain have been discussed for TPE such as multi-objective optimization and multi-fidelity optimization. However, the practical applications of HPO are not limited to DL, and black-box combinatorial optimization is actively utilized in some domains, e.g., chemistry and biology. As combinatorial optimization has been an untouched, yet very important, topic in TPE, we propose an efficient combinatorial optimization algorithm for TPE. In this paper, we first generalize the categorical kernel with the numerical kernel in TPE, enabling us to introduce a distance structure to the categorical kernel. Then we discuss modifications for the newly developed kernel to handle a large combinatorial search space. These modifications reduce the time complexity of the kernel calculation with respect to the size of a combinatorial search space. In the experiments using synthetic problems, we verified that our proposed method identifies better solutions with fewer evaluations than the original TPE. Our algorithm is available in Optuna, an open-source framework for HPO.

Modeling human behavior in urban environments is fundamental for social science, behavioral studies, and urban planning. Prior work often rely on rigid, hand-crafted rules, limiting their ability to simulate nuanced intentions, plans, and adaptive behaviors. Addressing these challenges, we envision an urban simulator (CitySim), capitalizing on breakthroughs in human-level intelligence exhibited by large language models. In CitySim, agents generate realistic daily schedules using a recursive value-driven approach that balances mandatory activities, personal habits, and situational factors. To enable long-term, lifelike simulations, we endow agents with beliefs, long-term goals, and spatial memory for navigation. CitySim exhibits closer alignment with real humans than prior work, both at micro and macro levels. Additionally, we conduct insightful experiments by modeling tens of thousands of agents and evaluating their collective behaviors under various real-world scenarios, including estimating crowd density, predicting place popularity, and assessing well-being. Our results highlight CitySim as a scalable, flexible testbed for understanding and forecasting urban phenomena.

Deep learning has recently revealed the existence of scaling laws, demonstrating that model performance follows predictable trends based on dataset and model sizes. Inspired by these findings and fascinating phenomena emerging in the over-parameterized regime, we examine a parallel direction: do similar scaling laws govern predictive uncertainties in deep learning? In identifiable parametric models, such scaling laws can be derived in a straightforward manner by treating model parameters in a Bayesian way. In this case, for example, we obtain $O(1/N)$ contraction rates for epistemic uncertainty with respect to the number of data $N$. However, in over-parameterized models, these guarantees do not hold, leading to largely unexplored behaviors. In this work, we empirically show the existence of scaling laws associated with various measures of predictive uncertainty with respect to dataset and model sizes. Through experiments on vision and language tasks, we observe such scaling laws for in- and out-of-distribution predictive uncertainty estimated through popular approximate Bayesian inference and ensemble methods. Besides the elegance of scaling laws and the practical utility of extrapolating uncertainties to larger data or models, this work provides strong evidence to dispel recurring skepticism against Bayesian approaches: "In many applications of deep learning we have so much data available: what do we need Bayes for?". Our findings show that "so much data" is typically not enough to make epistemic uncertainty negligible.

Speech foundation models (SFMs), such as Open Whisper-Style Speech Models (OWSM), are trained on massive datasets to achieve accurate automatic speech recognition. However, even SFMs struggle to accurately recognize rare and unseen words. While contextual biasing (CB) is a promising approach to improve recognition of such words, most CB methods are trained from scratch, resulting in lower performance than SFMs due to the lack of pre-trained knowledge. This paper integrates an existing CB method with OWSM v3.1 while freezing its pre-trained parameters. By leveraging the knowledge embedded in SFMs, the proposed method enables effective CB while preserving the advantages of SFMs, even with a small dataset. Experimental results show that the proposed method improves the biasing word error rate (B-WER) by 11.6 points, resulting in a 0.9 point improvement in the overall WER while reducing the real-time factor by 7.5% compared to the non-biasing baseline on the LibriSpeech 100 test-clean set.

Contextual biasing (CB) improves automatic speech recognition for rare and unseen phrases. Recent studies have introduced dynamic vocabulary, which represents context phrases as expandable tokens in autoregressive (AR) models. This method improves CB accuracy but with slow inference speed. While dynamic vocabulary can be applied to non-autoregressive (NAR) models, such as connectionist temporal classification (CTC), the conditional independence assumption fails to capture dependencies between static and dynamic tokens. This paper proposes DYNAC (Dynamic Vocabulary-based NAR Contextualization), a self-conditioned CTC method that integrates dynamic vocabulary into intermediate layers. Conditioning the encoder on dynamic vocabulary, DYNAC effectively captures dependencies between static and dynamic tokens while reducing the real-time factor (RTF). Experimental results show that DYNAC reduces RTF by 81% with a 0.1-point degradation in word error rate on the LibriSpeech 960 test-clean set.

Anyone who played with this toy will also remember the sound it made. Once you slid the power button to the On position, you heard a constant whirring sound of plastic gears turning and twisting. And if you tried to push it past its boundaries, it twitched and protested with a jarring "CLICK … CLICK … CLICK." It wasn't just kids who found the Armatron so special. It was featured on the cover of the November/December 1982 issue of Robotics Age magazine, which noted that the 31.95