Since its introduction in 2019, the whole end-to-end neural diarization (EEND) line of work has been addressing speaker diarization as a frame-wise multi-label classification problem with permutation-invariant training. Despite EEND showing great promise, a few recent works took a step back and studied the possible combination of (local) supervised EEND diarization with (global) unsupervised clustering. Yet, these hybrid contributions did not question the original multi-label formulation. We propose to switch from multi-label (where any two speakers can be active at the same time) to powerset multi-class classification (where dedicated classes are assigned to pairs of overlapping speakers). Through extensive experiments on 9 different benchmarks, we show that this formulation leads to significantly better performance (mostly on overlapping speech) and robustness to domain mismatch, while eliminating the detection threshold hyperparameter, critical for the multi-label formulation.

The real-world deployment of an autonomous driving system requires its components to run on-board and in real-time, including the motion prediction module that predicts the future trajectories of surrounding traffic participants. Existing agent-centric methods have demonstrated outstanding performance on public benchmarks. However, they suffer from high computational overhead and poor scalability as the number of agents to be predicted increases. To address this problem, we introduce the K-nearest neighbor attention with relative pose encoding (KNARPE), a novel attention mechanism allowing the pairwise-relative representation to be used by Transformers. Then, based on KNARPE we present the Heterogeneous Polyline Transformer with Relative pose encoding (HPTR), a hierarchical framework enabling asynchronous token update during the online inference. By sharing contexts among agents and reusing the unchanged contexts, our approach is as efficient as scene-centric methods, while performing on par with state-of-the-art agent-centric methods. Experiments on Waymo and Argoverse-2 datasets show that HPTR achieves superior performance among end-to-end methods that do not apply expensive post-processing or model ensembling. The code is available at https://github.com/zhejz/HPTR.

Stochastic gradient descent (SGD) and its variants are the main workhorses for solving large-scale optimization problems with nonconvex objective functions. Although the convergence of SGDs in the (strongly) convex case is well-understood, their convergence for nonconvex functions stands on weak mathematical foundations. Most existing studies on the nonconvex convergence of SGD show the complexity results based on either the minimum of the expected gradient norm or the functional sub-optimality gap (for functions with extra structural property) by searching the entire range of iterates. Hence the last iterations of SGDs do not necessarily maintain the same complexity guarantee. This paper shows that an $\epsilon$-stationary point exists in the final iterates of SGDs, given a large enough total iteration budget, $T$, not just anywhere in the entire range of iterates -- a much stronger result than the existing one. Additionally, our analyses allow us to measure the density of the $\epsilon$-stationary points in the final iterates of SGD, and we recover the classical $O(\frac{1}{\sqrt{T}})$ asymptotic rate under various existing assumptions on the objective function and the bounds on the stochastic gradient. As a result of our analyses, we addressed certain myths and legends related to the nonconvex convergence of SGD and posed some thought-provoking questions that could set new directions for research.

In this work, we study rapid, step-wise improvements of the loss in transformers when being confronted with multi-step decision tasks. We found that transformers struggle to learn the intermediate tasks, whereas CNNs have no such issue on the tasks we studied. When transformers learn the intermediate task, they do this rapidly and unexpectedly after both training and validation loss saturated for hundreds of epochs. We call these rapid improvements Eureka-moments, since the transformer appears to suddenly learn a previously incomprehensible task. Similar leaps in performance have become known as Grokking. In contrast to Grokking, for Eureka-moments, both the validation and the training loss saturate before rapidly improving. We trace the problem back to the Softmax function in the self-attention block of transformers and show ways to alleviate the problem. These fixes improve training speed. The improved models reach 95% of the baseline model in just 20% of training steps while having a much higher likelihood to learn the intermediate task, lead to higher final accuracy and are more robust to hyper-parameters.

We explain the methodology used to create the data submitted to HuMob Challenge, a data analysis competition for human mobility prediction. We adopted a personalized model to predict the individual's movement trajectory from their data, instead of predicting from the overall movement, based on the hypothesis that human movement is unique to each person. We devised the features such as the date and time, activity time, days of the week, time of day, and frequency of visits to POI (Point of Interest). As additional features, we incorporated the movement of other individuals with similar behavior patterns through the employment of clustering. The machine learning model we adopted was the Support Vector Regression (SVR). We performed accuracy through offline assessment and carried out feature selection and parameter tuning. Although overall dataset provided consists of 100,000 users trajectory, our method use only 20,000 target users data, and do not need to use other 80,000 data. Despite the personalized model's traditional feature engineering approach, this model yields reasonably good accuracy with lower computational cost.

Existing hierarchical forecasting techniques scale poorly when the number of time series increases. We propose to learn a coherent forecast for millions of time series with a single bottom-level forecast model by using a sparse loss function that directly optimizes the hierarchical product and/or temporal structure. The benefit of our sparse hierarchical loss function is that it provides practitioners a method of producing bottom-level forecasts that are coherent to any chosen cross-sectional or temporal hierarchy. In addition, removing the need for a post-processing step as required in traditional hierarchical forecasting techniques reduces the computational cost of the prediction phase in the forecasting pipeline. On the public M5 dataset, our sparse hierarchical loss function performs up to 10% (RMSE) better compared to the baseline loss function. We implement our sparse hierarchical loss function within an existing forecasting model at bol, a large European e-commerce platform, resulting in an improved forecasting performance of 2% at the product level. Finally, we found an increase in forecasting performance of about 5-10% when evaluating the forecasting performance across the cross-sectional hierarchies that we defined. These results demonstrate the usefulness of our sparse hierarchical loss applied to a production forecasting system at a major e-commerce platform.

Recently there has been a lot of interest in non-autoregressive (non-AR) models for speech synthesis, such as FastSpeech 2 and diffusion models. Unlike AR models, these models do not have autoregressive dependencies among outputs which makes inference efficient. This paper expands the range of available non-AR models with another member called energy-based models (EBMs). The paper describes how noise contrastive estimation, which relies on the comparison between positive and negative samples, can be used to train EBMs. It proposes a number of strategies for generating effective negative samples, including using high-performing AR models. It also describes how sampling from EBMs can be performed using Langevin Markov Chain Monte-Carlo (MCMC). The use of Langevin MCMC enables to draw connections between EBMs and currently popular diffusion models. Experiments on LJSpeech dataset show that the proposed approach offers improvements over Tacotron 2.

Classes of target functions containing a large number of approximately orthogonal elements are known to be hard to learn by the Statistical Query algorithms. Recently this classical fact re-emerged in a theory of gradient-based optimization of neural networks. In the novel framework, the hardness of a class is usually quantified by the variance of the gradient with respect to a random choice of a target function. A set of functions of the form $x\to ax \bmod p$, where $a$ is taken from ${\mathbb Z}_p$, has attracted some attention from deep learning theorists and cryptographers recently. This class can be understood as a subset of $p$-periodic functions on ${\mathbb Z}$ and is tightly connected with a class of high-frequency periodic functions on the real line. We present a mathematical analysis of limitations and challenges associated with using gradient-based learning techniques to train a high-frequency periodic function or modular multiplication from examples. We highlight that the variance of the gradient is negligibly small in both cases when either a frequency or the prime base $p$ is large. This in turn prevents such a learning algorithm from being successful.

Temporal data distribution shift is prevalent in the financial text. How can a financial sentiment analysis system be trained in a volatile market environment that can accurately infer sentiment and be robust to temporal data distribution shifts? In this paper, we conduct an empirical study on the financial sentiment analysis system under temporal data distribution shifts using a real-world financial social media dataset that spans three years. We find that the fine-tuned models suffer from general performance degradation in the presence of temporal distribution shifts. Furthermore, motivated by the unique temporal nature of the financial text, we propose a novel method that combines out-of-distribution detection with time series modeling for temporal financial sentiment analysis. Experimental results show that the proposed method enhances the model's capability to adapt to evolving temporal shifts in a volatile financial market.

The fast-paced development of machine learning (ML) methods coupled with its increasing adoption in research poses challenges for researchers without extensive training in ML. In neuroscience, for example, ML can help understand brain-behavior relationships, diagnose diseases, and develop biomarkers using various data sources like magnetic resonance imaging and electroencephalography. The primary objective of ML is to build models that can make accurate predictions on unseen data. Researchers aim to prove the existence of such generalizable models by evaluating performance using techniques such as cross-validation (CV), which uses systematic subsampling to estimate the generalization performance. Choosing a CV scheme and evaluating an ML pipeline can be challenging and, if used improperly, can lead to overestimated results and incorrect interpretations. We created julearn, an open-source Python library, that allow researchers to design and evaluate complex ML pipelines without encountering in common pitfalls. In this manuscript, we present the rationale behind julearn's design, its core features, and showcase three examples of previously-published research projects that can be easily implemented using this novel library. Julearn aims to simplify the entry into the ML world by providing an easy-to-use environment with built in guards against some of the most common ML pitfalls. With its design, unique features and simple interface, it poses as a useful Python-based library for research projects.