This paper describes several improvements to a new method for signal decomposition that we recently formulated under the name of Differentiable Dictionary Search (DDS). The fundamental idea of DDS is to exploit a class of powerful deep invertible density estimators called normalizing flows, to model the dictionary in a linear decomposition method such as NMF, effectively creating a bijection between the space of dictionary elements and the associated probability space, allowing a differentiable search through the dictionary space, guided by the estimated densities. As the initial formulation was a proof of concept with some practical limitations, we will present several steps towards making it scalable, hoping to improve both the computational complexity of the method and its signal decomposition capabilities. As a testbed for experimental evaluation, we choose the task of frame-level piano transcription, where the signal is to be decomposed into sources whose activity is attributed to individual piano notes. To highlight the impact of improved non-linear modelling of sources, we compare variants of our method to a linear overcomplete NMF baseline. Experimental results will show that even in the absence of additional constraints, our models produce increasingly sparse and precise decompositions, according to two pertinent evaluation measures.

Our system is motivated by the limitations of current frameworks, which restrict the dynamic routing in MoE layers to satisfy the constraints of existing software and hardware. These formulations force a tradeoff between model quality and hardware efficiency, as users must choose between dropping tokens from the computation or wasting computation and memory on padding. To address these limitations, we reformulate MoE computation in terms of block-sparse operations and develop new block-sparse GPU kernels that efficiently handle the dynamism present in MoEs. Our approach never drops tokens and maps efficiently to modern hardware, enabling end-to-end training speedups of up to 40% over MoEs trained with the state-of-the-art Tutel library and 2.4 over DNNs trained with the highly-optimized Megatron-LM framework. The past decade has seen significant progress in are fundamental to these architectures. However, existing algorithms and high-performance software to make sparsity hardware and software for deep learning make it difficult practically useful (Gray et al., 2017; Narang et al., 2017; to meet this challenge. For example, TPUs and their XLA Kalchbrenner et al., 2018; Elsen et al., 2020; Gale et al., compiler require all tensor shapes to be known statically 2020).

This paper introduced key aspects of applying Machine Learning (ML) models, improved trading strategies, and the Quasi-Reversibility Method (QRM) to optimize stock option forecasting and trading results. It presented the findings of the follow-up project of the research "Application of Convolutional Neural Networks with Quasi-Reversibility Method Results for Option Forecasting". First, the project included an application of Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks to provide a novel way of predicting stock option trends. Additionally, it examined the dependence of the ML models by evaluating the experimental method of combining multiple ML models to improve prediction results and decision-making. Lastly, two improved trading strategies and simulated investing results were presented. The Binomial Asset Pricing Model with discrete time stochastic process analysis and portfolio hedging was applied and suggested an optimized investment expectation. These results can be utilized in real-life trading strategies to optimize stock option investment results based on historical data.

We introduce a novel way to incorporate prior information into (semi-) supervised non-negative matrix factorization, which we call differentiable dictionary search. It enables general, highly flexible and principled modelling of mixtures where non-linear sources are linearly mixed. We study its behavior on an audio decomposition task, and conduct an extensive, highly controlled study of its modelling capabilities.

Recently, a new distributed learning scheme called Federated Learning (FL) has been introduced. FL is designed so that server never collects user-owned data meaning it is great at preserving privacy. FL's process starts with the server sending a model to clients, then the clients train that model using their data and send the updated model back to the server. Afterward, the server aggregates all the updates and modifies the global model. This process is repeated until the model converges. This study focuses on an FL setting called cross-device FL, which trains based on a large number of clients. Since many devices may be unavailable in cross-device FL, and communication between the server and all clients is extremely costly, only a fraction of clients gets selected for training at each round. In vanilla FL, clients are selected randomly, which results in an acceptable accuracy but is not ideal from the overall training time perspective, since some clients are slow and can cause some training rounds to be slow. If only fast clients get selected the learning would speed up, but it will be biased toward only the fast clients' data, and the accuracy degrades. Consequently, new client selection techniques have been proposed to improve the training time by considering individual clients' resources and speed. This paper introduces the first availability-aware selection strategy called MDA. The results show that our approach makes learning faster than vanilla FL by up to 6.5%. Moreover, we show that resource heterogeneity-aware techniques are effective but can become even better when combined with our approach, making it faster than the state-of-the-art selectors by up to 16%. Lastly, our approach selects more unique clients for training compared to client selectors that only select fast clients, which reduces our technique's bias.

Semantic communication is viewed as a revolutionary paradigm that can potentially transform how we design and operate wireless communication systems. However, despite a recent surge of research activities in this area, the research landscape remains limited. In this tutorial, we present the first rigorous vision of a scalable end-to-end semantic communication network that is founded on novel concepts from artificial intelligence (AI), causal reasoning, and communication theory. We first discuss how the design of semantic communication networks requires a move from data-driven networks towards knowledge-driven ones. Subsequently, we highlight the necessity of creating semantic representations of data that satisfy the key properties of minimalism, generalizability, and efficiency so as to do more with less. We then explain how those representations can form the basis a so-called semantic language. By using semantic representation and languages, we show that the traditional transmitter and receiver now become a teacher and apprentice. Then, we define the concept of reasoning by investigating the fundamentals of causal representation learning and their role in designing semantic communication networks. We demonstrate that reasoning faculties are majorly characterized by the ability to capture causal and associational relationships in datastreams. For such reasoning-driven networks, we propose novel and essential semantic communication metrics that include new "reasoning capacity" measures that could go beyond Shannon's bound to capture the convergence of computing and communication. Finally, we explain how semantic communications can be scaled to large-scale networks (6G and beyond). In a nutshell, we expect this tutorial to provide a comprehensive reference on how to properly build, analyze, and deploy future semantic communication networks.

Current language models achieve low perplexity but their resulting generations still suffer from toxic responses, repetitiveness and contradictions. The standard language modeling setup fails to address these issues. In this paper, we introduce a new architecture, {\sc Director}, that consists of a unified generator-classifier with both a language modeling and a classification head for each output token. Training is conducted jointly using both standard language modeling data, and data labeled with desirable and undesirable sequences. Experiments in several settings show that the model has competitive training and decoding speed compared to standard language models while yielding superior results, alleviating known issues while maintaining generation quality. It also outperforms existing model guiding approaches in terms of both accuracy and efficiency.

Humans experience small fluctuations in their gait when walking on uneven terrain. The fluctuations deviate from the steady, energy-minimizing pattern for level walking, and have no obvious organization. But humans often look ahead when they walk, and could potentially plan anticipatory fluctuations for the terrain. Such planning is only sensible if it serves some an objective purpose, such as maintaining constant speed or reducing energy expenditure, that is also attainable within finite planning capacity. Here we show that humans do plan and perform optimal control strategies on uneven terrain. Rather than maintain constant speed, they make purposeful, anticipatory speed adjustments that are consistent with minimizing energy expenditure. A simple optimal control model predicts economical speed fluctuations that agree well with experiments with humans (N = 12) walking on seven different terrain profiles (correlated with model ro=0.55+-0.11, P<0.05 all terrains). Participants made repeatable speed fluctuations starting about six to eight steps ahead of each terrain feature (up to +-7.5 cm height difference each step, up to 16 consecutive features). Nearer features matter more, because energy is dissipated with each succeeding step collision with ground, preventing momentum from persisting indefinitely. A finite horizon of continuous look ahead and motor working space thus suffice to practically optimize for any length of terrain. Humans reason about walking in the near future to plan complex optimal control sequences.

Dimensional reduction~(DR) maps high-dimensional data into a lower dimensions latent space with minimized defined optimization objectives. The DR method usually falls into feature selection~(FS) and feature projection~(FP). FS focuses on selecting a critical subset of dimensions but risks destroying the data distribution (structure). On the other hand, FP combines all the input features into lower dimensions space, aiming to maintain the data structure; but lacks interpretability and sparsity. FS and FP are traditionally incompatible categories; thus, they have not been unified into an amicable framework. We propose that the ideal DR approach combines both FS and FP into a unified end-to-end manifold learning framework, simultaneously performing fundamental feature discovery while maintaining the intrinsic relationships between data samples in the latent space. In this work, we develop a unified framework, Unified Dimensional Reduction Neural-network~(UDRN), that integrates FS and FP in a compatible, end-to-end way. We improve the neural network structure by implementing FS and FP tasks separately using two stacked sub-networks. In addition, we designed data augmentation of the DR process to improve the generalization ability of the method when dealing with extensive feature datasets and designed loss functions that can cooperate with the data augmentation. Extensive experimental results on four image and four biological datasets, including very high-dimensional data, demonstrate the advantages of DRN over existing methods~(FS, FP, and FS\&FP pipeline), especially in downstream tasks such as classification and visualization.

Emotion recognition is a key attribute for artificial intelligence systems that need to naturally interact with humans. However, the task definition is still an open problem due to the inherent ambiguity of emotions. In this paper, a novel Bayesian training loss based on per-utterance Dirichlet prior distributions is proposed for verbal emotion recognition, which models the uncertainty in one-hot labels created when human annotators assign the same utterance to different emotion classes. An additional metric is used to evaluate the performance by detection test utterances with high labelling uncertainty. This removes a major limitation that emotion classification systems only consider utterances with labels where the majority of annotators agree on the emotion class. Furthermore, a frequentist approach is studied to leverage the continuous-valued "soft" labels obtained by averaging the one-hot labels. We propose a two-branch model structure for emotion classification on a per-utterance basis, which achieves state-of-the-art classification results on the widely used IEMOCAP dataset. Based on this, uncertainty estimation experiments were performed. The best performance in terms of the area under the precision-recall curve when detecting utterances with high uncertainty was achieved by interpolating the Bayesian training loss with the Kullback-Leibler divergence training loss for the soft labels. The generality of the proposed approach was verified using the MSP-Podcast dataset which yielded the same pattern of results.