Machine Learning (cs.LG)

Abstract: We study the problem of maintaining a differentially private decaying sum under continual observation. We give a unifying framework and an efficient algorithm for this problem for \emph{any sufficiently smooth} function. Our algorithm is the first differentially private algorithm that does not have a multiplicative error for polynomially-decaying weights. Our algorithm improves on all prior works on differentially private decaying sums under continual observation and recovers exactly the additive error for the special case of continual counting from Henzinger et al. (SODA 2023) as a corollary. Our algorithm is a variant of the factorization mechanism whose error depends on the $\gamma_2$ and $\gamma_F$ norm of the underlying matrix. We give a constructive proof for an almost exact upper bound on the $\gamma_2$ and $\gamma_F$ norm and an almost tight lower bound on the $\gamma_2$ norm for a large class of lower-triangular matrices. This is the first non-trivial lower bound for lower-triangular matrices whose non-zero entries are not all the same. It includes matrices for all continual decaying sums problems, resulting in an upper bound on the additive error of any differentially private decaying sums algorithm under continual observation. We also explore some implications of our result in discrepancy theory and operator algebra. Given the importance of the $\gamma_2$ norm in computer science and the extensive work in mathematics, we believe our result will have further applications.

Abstract: Neural networks have achieved remarkable performance in various application domains. Nevertheless, a large number of weights in pre-trained deep neural networks prohibit them from being deployed on smartphones and embedded systems. It is highly desirable to obtain lightweight versions of neural networks for inference in edge devices. Many cost-effective approaches were proposed to prune dense and convolutional layers that are common in deep neural networks and dominant in the parameter space. However, a unified theoretical foundation for the problem mostly is missing. In this paper, we identify the close connection between matrix spectrum learning and neural network training for dense and convolutional layers and argue that weight pruning is essentially a matrix sparsification process to preserve the spectrum. Based on the analysis, we also propose a matrix sparsification algorithm tailored for neural network pruning that yields better pruning result. We carefully design and conduct experiments to support our arguments. Hence we provide a consolidated viewpoint for neural network pruning and enhance the interpretability of deep neural networks by identifying and preserving the critical neural weights.

Abstract: Drug-target binding affinity prediction plays an important role in the early stages of drug discovery, which can infer the strength of interactions between new drugs and new targets. However, the performance of previous computational models is limited by the following drawbacks. The learning of drug representation relies only on supervised data, without taking into account the information contained in the molecular graph itself. Moreover, most previous studies tended to design complicated representation learning module, while uniformity, which is used to measure representation quality, is ignored. In this study, we propose GraphCL-DTA, a graph contrastive learning with molecular semantics for drug-target binding affinity prediction. In GraphCL-DTA, we design a graph contrastive learning framework for molecular graphs to learn drug representations, so that the semantics of molecular graphs are preserved. Through this graph contrastive framework, a more essential and effective drug representation can be learned without additional supervised data. Next, we design a new loss function that can be directly used to smoothly adjust the uniformity of drug and target representations. By directly optimizing the uniformity of representations, the representation quality of drugs and targets can be improved. The effectiveness of the above innovative elements is verified on two real datasets, KIBA and Davis. The excellent performance of GraphCL-DTA on the above datasets suggests its superiority to the state-of-the-art model.

Abstract: A recent line of work has shown a surprising connection between multicalibration, a multi-group fairness notion, and omniprediction, a learning paradigm that provides simultaneous loss minimization guarantees for a large family of loss functions. Prior work studies omniprediction in the batch setting. We initiate the study of omniprediction in the online adversarial setting. Although there exist algorithms for obtaining notions of multicalibration in the online adversarial setting, unlike batch algorithms, they work only for small finite classes of benchmark functions $F$, because they require enumerating every function $f \in F$ at every round. In contrast, omniprediction is most interesting for learning theoretic hypothesis classes $F$, which are generally continuously large. We develop a new online multicalibration algorithm that is well defined for infinite benchmark classes $F$, and is oracle efficient (i.e. for any class $F$, the algorithm has the form of an efficient reduction to a no-regret learning algorithm for $F$). The result is the first efficient online omnipredictor -- an oracle efficient prediction algorithm that can be used to simultaneously obtain no regret guarantees to all Lipschitz convex loss functions. For the class $F$ of linear functions, we show how to make our algorithm efficient in the worst case. Also, we show upper and lower bounds on the extent to which our rates can be improved: our oracle efficient algorithm actually promises a stronger guarantee called swap-omniprediction, and we prove a lower bound showing that obtaining $O(\sqrt{T})$ bounds for swap-omniprediction is impossible in the online setting. On the other hand, we give a (non-oracle efficient) algorithm which can obtain the optimal $O(\sqrt{T})$ omniprediction bounds without going through multicalibration, giving an information theoretic separation between these two solution concepts.

Abstract: Time series prediction plays a crucial role in various industrial fields. In recent years, neural networks with a transformer backbone have achieved remarkable success in many domains, including computer vision and NLP. In time series analysis domain, some studies have suggested that even the simplest MLP networks outperform advanced transformer-based networks on time series forecast tasks. However, we believe these findings indicate there to be low-rank properties in time series sequences. In this paper, we consider the low-pass characteristics of transformers and try to incorporate the advantages of MLP. We adopt skip-layer connections inspired by Unet into traditional transformer backbone, thus preserving high-frequency context from input to output, namely U-shaped Transformer. We introduce patch merge and split operation to extract features with different scales and use larger datasets to fully make use of the transformer backbone. Our experiments demonstrate that the model performs at an advanced level across multiple datasets with relatively low cost.

Abstract: Solving partial differential equations (PDEs) using a data-driven approach has become increasingly common. The recent development of the operator learning paradigm has enabled the solution of a broader range of PDE-related problems. We propose an operator learning method to solve time-dependent PDEs continuously in time without needing any temporal discretization. The proposed approach, named DiTTO, is inspired by latent diffusion models. While diffusion models are usually used in generative artificial intelligence tasks, their time-conditioning mechanism is extremely useful for PDEs. The diffusion-inspired framework is combined with elements from the Transformer architecture to improve its capabilities. We demonstrate the effectiveness of the new approach on a wide variety of PDEs in multiple dimensions, namely the 1-D Burgers' equation, 2-D Navier-Stokes equations, and the acoustic wave equation in 2-D and 3-D. DiTTO achieves state-of-the-art results in terms of accuracy for these problems. We also present a method to improve the performance of DiTTO by using fast sampling concepts from diffusion models. Finally, we show that DiTTO can accurately perform zero-shot super-resolution in time.

Abstract: Energy shortfall and electricity load shedding are the main problems for developing countries. The main causes are lack of management in the energy sector and the use of non-renewable energy sources. The improved energy management and use of renewable sources can be significant to resolve energy crisis. It is necessary to increase the use of renewable energy sources (RESs) to meet the increasing energy demand due to high prices of fossil-fuel based energy. Federated learning (FL) is the most emerging technique in the field of artificial intelligence. Federated learning helps to generate global model at server side by ensemble locally trained models at remote edges sites while preserving data privacy. The global model used to predict energy demand to satisfy the needs of consumers. In this article, we have proposed Blockchain based safe distributed ledger technology for transaction of data between prosumer and consumer to ensure their transparency, traceability and security. Furthermore, we have also proposed a Federated learning model to forecast the energy requirements of consumer and prosumer. Moreover, Blockchain has been used to store excess energy data from prosumer for better management of energy between prosumer and grid. Lastly, the experiment results revealed that renewable energy sources have produced better and comparable results to other non-renewable energy resources.

Abstract: Sequential decision-making under uncertainty is often associated with long feedback delays. Such delays degrade the performance of the learning agent in identifying a subset of arms with the optimal collective reward in the long run. This problem becomes significantly challenging in a non-stationary environment with structural dependencies amongst the reward distributions associated with the arms. Therefore, besides adapting to delays and environmental changes, learning the causal relations alleviates the adverse effects of feedback delay on the decision-making process. We formalize the described setting as a non-stationary and delayed combinatorial semi-bandit problem with causally related rewards. We model the causal relations by a directed graph in a stationary structural equation model. The agent maximizes the long-term average payoff, defined as a linear function of the base arms' rewards. We develop a policy that learns the structural dependencies from delayed feedback and utilizes that to optimize the decision-making while adapting to drifts. We prove a regret bound for the performance of the proposed algorithm. Besides, we evaluate our method via numerical analysis using synthetic and real-world datasets to detect the regions that contribute the most to the spread of Covid-19 in Italy.

Abstract: Several Active Learning (AL) policies require retraining a target model several times in order to identify the most informative samples and rarely offer the option to focus on the acquisition of samples from underrepresented classes. Here the Mining of Single-Class by Active Learning (MiSiCAL) paradigm is introduced where an AL policy is constructed through deep reinforcement learning and exploits quantity-accuracy correlations to build datasets on which high-performance models can be trained with regards to specific classes. MiSiCAL is especially helpful in the case of very large batch sizes since it does not require repeated model training sessions as is common in other AL methods. This is thanks to its ability to exploit fixed representations of the candidate data points. We find that MiSiCAL is able to outperform a random policy on 150 out of 171 COCO10k classes, while the strongest baseline only outperforms random on 101 classes.

Abstract: Efficiency and trustworthiness are two eternal pursuits when applying deep learning in real-world applications. With regard to efficiency, dataset distillation (DD) endeavors to reduce training costs by distilling the large dataset into a tiny synthetic dataset. However, existing methods merely concentrate on in-distribution (InD) classification in a closed-world setting, disregarding out-of-distribution (OOD) samples. On the other hand, OOD detection aims to enhance models' trustworthiness, which is always inefficiently achieved in full-data settings. For the first time, we simultaneously consider both issues and propose a novel paradigm called Trustworthy Dataset Distillation (TrustDD). By distilling both InD samples and outliers, the condensed datasets are capable to train models competent in both InD classification and OOD detection. To alleviate the requirement of real outlier data and make OOD detection more practical, we further propose to corrupt InD samples to generate pseudo-outliers and introduce Pseudo-Outlier Exposure (POE). Comprehensive experiments on various settings demonstrate the effectiveness of TrustDD, and the proposed POE surpasses state-of-the-art method Outlier Exposure (OE). Compared with the preceding DD, TrustDD is more trustworthy and applicable to real open-world scenarios. Our code will be publicly available.

Abstract: With an increasing number of smart devices like internet of things (IoT) devices deployed in the field, offloadingtraining of neural networks (NNs) to a central server becomes more and more infeasible. Recent efforts toimprove users' privacy have led to on-device learning emerging as an alternative. However, a model trainedonly on a single device, using only local data, is unlikely to reach a high accuracy. Federated learning (FL)has been introduced as a solution, offering a privacy-preserving trade-off between communication overheadand model accuracy by sharing knowledge between devices but disclosing the devices' private data. Theapplicability and the benefit of applying baseline FL are, however, limited in many relevant use cases dueto the heterogeneity present in such environments. In this survey, we outline the heterogeneity challengesFL has to overcome to be widely applicable in real-world applications. We especially focus on the aspect ofcomputation heterogeneity among the participating devices and provide a comprehensive overview of recentworks on heterogeneity-aware FL. We discuss two groups: works that adapt the NN architecture and worksthat approach heterogeneity on a system level, covering Federated Averaging (FedAvg), distillation, and splitlearning-based approaches, as well as synchronous and asynchronous aggregation schemes.

Abstract: In many reinforcement learning tasks, the agent has to learn to interact with many objects of different types and generalize to unseen combinations and numbers of objects. Often a task is a composition of previously learned tasks (e.g. block stacking). These are examples of compositional generalization, in which we compose object-centric representations to solve complex tasks. Recent works have shown the benefits of object-factored representations and hierarchical abstractions for improving sample efficiency in these settings. On the other hand, these methods do not fully exploit the benefits of factorization in terms of object attributes. In this paper, we address this opportunity and introduce the Dynamic Attribute FacTored RL (DAFT-RL) framework. In DAFT-RL, we leverage object-centric representation learning to extract objects from visual inputs. We learn to classify them in classes and infer their latent parameters. For each class of object, we learn a class template graph that describes how the dynamics and reward of an object of this class factorize according to its attributes. We also learn an interaction pattern graph that describes how objects of different classes interact with each other at the attribute level. Through these graphs and a dynamic interaction graph that models the interactions between objects, we can learn a policy that can then be directly applied in a new environment by just estimating the interactions and latent parameters. We evaluate DAFT-RL in three benchmark datasets and show our framework outperforms the state-of-the-art in generalizing across unseen objects with varying attributes and latent parameters, as well as in the composition of previously learned tasks.

Abstract: We study the size of a neural network needed to approximate the maximum function over $d$ inputs, in the most basic setting of approximating with respect to the $L_2$ norm, for continuous distributions, for a network that uses ReLU activations. We provide new lower and upper bounds on the width required for approximation across various depths. Our results establish new depth separations between depth 2 and 3, and depth 3 and 5 networks, as well as providing a depth $\mathcal{O}(\log(\log(d)))$ and width $\mathcal{O}(d)$ construction which approximates the maximum function, significantly improving upon the depth requirements of the best previously known bounds for networks with linearly-bounded width. Our depth separation results are facilitated by a new lower bound for depth 2 networks approximating the maximum function over the uniform distribution, assuming an exponential upper bound on the size of the weights. Furthermore, we are able to use this depth 2 lower bound to provide tight bounds on the number of neurons needed to approximate the maximum by a depth 3 network. Our lower bounds are of potentially broad interest as they apply to the widely studied and used \emph{max} function, in contrast to many previous results that base their bounds on specially constructed or pathological functions and distributions.

Abstract: In this work we perform a scoping review of the current literature on the detection of throat cancer from speech recordings using machine learning and artificial intelligence. We find 22 papers within this area and discuss their methods and results. We split these papers into two groups - nine performing binary classification, and 13 performing multi-class classification. The papers present a range of methods with neural networks being most commonly implemented. Many features are also extracted from the audio before classification, with the most common bring mel-frequency cepstral coefficients. None of the papers found in this search have associated code repositories and as such are not reproducible. Therefore, we create a publicly available code repository of our own classifiers. We use transfer learning on a multi-class problem, classifying three pathologies and healthy controls. Using this technique we achieve an unweighted average recall of 53.54%, sensitivity of 83.14%, and specificity of 64.00%. We compare our classifiers with the results obtained on the same dataset and find similar results.

Abstract: Non-intrusive load monitoring (NILM) is the process of obtaining appliance-level data from a single metering point, measuring total electricity consumption of a household or a business. Appliance-level data can be directly used for demand response applications and energy management systems as well as for awareness raising and motivation for improvements in energy efficiency and reduction in the carbon footprint. Recently, classical machine learning and deep learning (DL) techniques became very popular and proved as highly effective for NILM classification, but with the growing complexity these methods are faced with significant computational and energy demands during both their training and operation. In this paper, we introduce a novel DL model aimed at enhanced multi-label classification of NILM with improved computation and energy efficiency. We also propose a testing methodology for comparison of different models using data synthesized from the measurement datasets so as to better represent real-world scenarios. Compared to the state-of-the-art, the proposed model has its carbon footprint reduced by more than 23% while providing on average approximately 8 percentage points in performance improvement when testing on data derived from REFIT and UK-DALE datasets.

Abstract: Nowadays, wind energy has drawn increasing attention as its important role in carbon neutrality and sustainable development. When wind power is integrated into the power grid, precise forecasting is necessary for the sustainability and security of the system. However, the unpredictable nature and long sequence prediction make it especially challenging. In this technical report, we introduce the BERT model applied for Baidu KDD Cup 2022, and the daily fluctuation is added by post-processing to make the predicted results in line with daily periodicity. Our solution achieves 3rd place of 2490 teams. The code is released athttps://github.com/LongxingTan/KDD2022-Baidu

Abstract: Recent advancements in Natural Language Processing (NLP) have witnessed the groundbreaking impact of pretrained models, yielding impressive outcomes across various tasks. This study seeks to extend the power of pretraining methodologies to tabular data, a domain traditionally overlooked, yet inherently challenging due to the plethora of table schemas intrinsic to different tasks. The primary research questions underpinning this work revolve around the adaptation to heterogeneous table structures, the establishment of a universal pretraining protocol for tabular data, the generalizability and transferability of learned knowledge across tasks, the adaptation to diverse downstream applications, and the incorporation of incremental columns over time. In response to these challenges, we introduce UniTabE, a pioneering method designed to process tables in a uniform manner, devoid of constraints imposed by specific table structures. UniTabE's core concept relies on representing each basic table element with a module, termed TabUnit. This is subsequently followed by a Transformer encoder to refine the representation. Moreover, our model is designed to facilitate pretraining and finetuning through the utilization of free-form prompts. In order to implement the pretraining phase, we curated an expansive tabular dataset comprising approximately 13 billion samples, meticulously gathered from the Kaggle platform. Rigorous experimental testing and analyses were performed under a myriad of scenarios to validate the effectiveness of our methodology. The experimental results demonstrate UniTabE's superior performance against several baseline models across a multitude of benchmark datasets. This, therefore, underscores UniTabE's potential to significantly enhance the semantic representation of tabular data, thereby marking a significant stride in the field of tabular data analysis.

Abstract: Uncertainty learning and quantification of models are crucial tasks to enhance the trustworthiness of the models. Importantly, the recent surge of generative language models (GLMs) emphasizes the need for reliable uncertainty quantification due to the concerns on generating hallucinated facts. In this paper, we propose to learn neural prediction set models that comes with the probably approximately correct (PAC) guarantee for quantifying the uncertainty of GLMs. Unlike existing prediction set models, which are parameterized by a scalar value, we propose to parameterize prediction sets via neural networks, which achieves more precise uncertainty quantification but still satisfies the PAC guarantee. We demonstrate the efficacy of our method on four types of language datasets and six types of models by showing that our method improves the quantified uncertainty by $63\%$ on average, compared to a standard baseline method.

Abstract: Machine learning for point clouds has been attracting much attention, with many applications in various fields, such as shape recognition and material science. To enhance the accuracy of such machine learning methods, it is known to be effective to incorporate global topological features, which are typically extracted by persistent homology. In the calculation of persistent homology for a point cloud, we need to choose a filtration for the point clouds, an increasing sequence of spaces. Because the performance of machine learning methods combined with persistent homology is highly affected by the choice of a filtration, we need to tune it depending on data and tasks. In this paper, we propose a framework that learns a filtration adaptively with the use of neural networks. In order to make the resulting persistent homology isometry-invariant, we develop a neural network architecture with such invariance. Additionally, we theoretically show a finite-dimensional approximation result that justifies our architecture. Experimental results demonstrated the efficacy of our framework in several classification tasks.

Abstract: Hyperbox-based classification has been seen as a promising technique in which decisions on the data are represented as a series of orthogonal, multidimensional boxes (i.e., hyperboxes) that are often interpretable and human-readable. However, existing methods are no longer capable of efficiently handling the increasing volume of data many application domains face nowadays. We address this gap by proposing a novel, fully differentiable framework for hyperbox-based classification via neural networks. In contrast to previous work, our hyperbox models can be efficiently trained in an end-to-end fashion, which leads to significantly reduced training times and superior classification results.

Abstract: In safety-critical classification tasks, conformal prediction allows to perform rigorous uncertainty quantification by providing confidence sets including the true class with a user-specified probability. This generally assumes the availability of a held-out calibration set with access to ground truth labels. Unfortunately, in many domains, such labels are difficult to obtain and usually approximated by aggregating expert opinions. In fact, this holds true for almost all datasets, including well-known ones such as CIFAR and ImageNet. Applying conformal prediction using such labels underestimates uncertainty. Indeed, when expert opinions are not resolvable, there is inherent ambiguity present in the labels. That is, we do not have ``crisp'', definitive ground truth labels and this uncertainty should be taken into account during calibration. In this paper, we develop a conformal prediction framework for such ambiguous ground truth settings which relies on an approximation of the underlying posterior distribution of labels given inputs. We demonstrate our methodology on synthetic and real datasets, including a case study of skin condition classification in dermatology.

Abstract: A neural solver and differentiable simulation of the quantum transmitting boundary model is presented for the inverse quantum transport problem. The neural solver is used to engineer continuous transmission properties and the differentiable simulation is used to engineer current-voltage characteristics.

Abstract: Traditional approaches for learning on categorical data underexploit the dependencies between columns (\aka fields) in a dataset because they rely on the embedding of data points driven alone by the classification/regression loss. In contrast, we propose a novel method for learning on categorical data with the goal of exploiting dependencies between fields. Instead of modelling statistics of features globally (i.e., by the covariance matrix of features), we learn a global field dependency matrix that captures dependencies between fields and then we refine the global field dependency matrix at the instance-wise level with different weights (so-called local dependency modelling) w.r.t. each field to improve the modelling of the field dependencies. Our algorithm exploits the meta-learning paradigm, i.e., the dependency matrices are refined in the inner loop of the meta-learning algorithm without the use of labels, whereas the outer loop intertwines the updates of the embedding matrix (the matrix performing projection) and global dependency matrix in a supervised fashion (with the use of labels). Our method is simple yet it outperforms several state-of-the-art methods on six popular dataset benchmarks. Detailed ablation studies provide additional insights into our method.

Abstract: Zero-cost proxies are nowadays frequently studied and used to search for neural architectures. They show an impressive ability to predict the performance of architectures by making use of their untrained weights. These techniques allow for immense search speed-ups. So far the joint search for well-performing and robust architectures has received much less attention in the field of NAS. Therefore, the main focus of zero-cost proxies is the clean accuracy of architectures, whereas the model robustness should play an evenly important part. In this paper, we analyze the ability of common zero-cost proxies to serve as performance predictors for robustness in the popular NAS-Bench-201 search space. We are interested in the single prediction task for robustness and the joint multi-objective of clean and robust accuracy. We further analyze the feature importance of the proxies and show that predicting the robustness makes the prediction task from existing zero-cost proxies more challenging. As a result, the joint consideration of several proxies becomes necessary to predict a model's robustness while the clean accuracy can be regressed from a single such feature.

Abstract: We consider the problem of learning a sparse graph underlying an undirected Gaussian graphical model, a key problem in statistical machine learning. Given $n$ samples from a multivariate Gaussian distribution with $p$ variables, the goal is to estimate the $p \times p$ inverse covariance matrix (aka precision matrix), assuming it is sparse (i.e., has a few nonzero entries). We propose GraphL0BnB, a new estimator based on an $\ell_0$-penalized version of the pseudolikelihood function, while most earlier approaches are based on the $\ell_1$-relaxation. Our estimator can be formulated as a convex mixed integer program (MIP) which can be difficult to compute at scale using off-the-shelf commercial solvers. To solve the MIP, we propose a custom nonlinear branch-and-bound (BnB) framework that solves node relaxations with tailored first-order methods. As a by-product of our BnB framework, we propose large-scale solvers for obtaining good primal solutions that are of independent interest. We derive novel statistical guarantees (estimation and variable selection) for our estimator and discuss how our approach improves upon existing estimators. Our numerical experiments on real/synthetic datasets suggest that our method can solve, to near-optimality, problem instances with $p = 10^4$ -- corresponding to a symmetric matrix of size $p \times p$ with $p^2/2$ binary variables. We demonstrate the usefulness of GraphL0BnB versus various state-of-the-art approaches on a range of datasets.

Abstract: Support Vector Machines (SVM) have gathered significant acclaim as classifiers due to their successful implementation of Statistical Learning Theory. However, in the context of multiclass and multilabel settings, the reliance on vector-based formulations in existing SVM-based models poses limitations regarding flexibility and ease of incorporating additional terms to handle specific challenges. To overcome these limitations, our research paper focuses on introducing a matrix formulation for SVM that effectively addresses these constraints. By employing the Accelerated Gradient Descent method in the dual, we notably enhance the efficiency of solving the Matrix-SVM problem. Experimental evaluations on multilabel and multiclass datasets demonstrate that Matrix SVM achieves superior time efficacy while delivering similar results to Binary Relevance SVM. Moreover, our matrix formulation unveils crucial insights and advantages that may not be readily apparent in traditional vector-based notations. We emphasize that numerous multilabel models can be viewed as extensions of SVM, with customised modifications to meet specific requirements. The matrix formulation presented in this paper establishes a solid foundation for developing more sophisticated models capable of effectively addressing the distinctive challenges encountered in multilabel learning.

Abstract: The use of machine learning in algorithmic trading systems is increasingly common. In a typical set-up, supervised learning is used to predict the future prices of assets, and those predictions drive a simple trading and execution strategy. This is quite effective when the predictions have sufficient signal, markets are liquid, and transaction costs are low. However, those conditions often do not hold in thinly traded financial markets and markets for differentiated assets such as real estate or vehicles. In these markets, the trading strategy must consider the long-term effects of taking positions that are relatively more difficult to change. In this work, we propose a Reinforcement Learning (RL) algorithm that trades based on signals from a learned predictive model and addresses these challenges. We test our algorithm on 20+ years of equity data from Bursa Malaysia.

Abstract: Maximizing long-term rewards is the primary goal in sequential decision-making problems. The majority of existing methods assume that side information is freely available, enabling the learning agent to observe all features' states before making a decision. In real-world problems, however, collecting beneficial information is often costly. That implies that, besides individual arms' reward, learning the observations of the features' states is essential to improve the decision-making strategy. The problem is aggravated in a non-stationary environment where reward and cost distributions undergo abrupt changes over time. To address the aforementioned dual learning problem, we extend the contextual bandit setting and allow the agent to observe subsets of features' states. The objective is to maximize the long-term average gain, which is the difference between the accumulated rewards and the paid costs on average. Therefore, the agent faces a trade-off between minimizing the cost of information acquisition and possibly improving the decision-making process using the obtained information. To this end, we develop an algorithm that guarantees a sublinear regret in time. Numerical results demonstrate the superiority of our proposed policy in a real-world scenario.

Abstract: Imitation Learning (IL) is one of the most widely used methods in machine learning. Yet, while powerful, many works find it is often not able to fully recover the underlying expert behavior. However, none of these works deeply investigate the role of scaling up the model and data size. Inspired by recent work in Natural Language Processing (NLP) where "scaling up" has resulted in increasingly more capable LLMs, we investigate whether carefully scaling up model and data size can bring similar improvements in the imitation learning setting. To demonstrate our findings, we focus on the game of NetHack, a challenging environment featuring procedural generation, stochasticity, long-term dependencies, and partial observability. We find IL loss and mean return scale smoothly with the compute budget and are strongly correlated, resulting in power laws for training compute-optimal IL agents with respect to model size and number of samples. We forecast and train several NetHack agents with IL and find they outperform prior state-of-the-art by at least 2x in all settings. Our work both demonstrates the scaling behavior of imitation learning in a challenging domain, as well as the viability of scaling up current approaches for increasingly capable agents in NetHack, a game that remains elusively hard for current AI systems.

Abstract: Extracting object-level representations for downstream reasoning tasks is an emerging area in AI. Learning object-centric representations in an unsupervised setting presents multiple challenges, a key one being binding an arbitrary number of object instances to a specialized object slot. Recent object-centric representation methods like Slot Attention utilize iterative attention to learn composable representations with dynamic inference level binding but fail to achieve specialized slot level binding. To address this, in this paper we propose Unsupervised Conditional Slot Attention using a novel Probabilistic Slot Dictionary (PSD). We define PSD with (i) abstract object-level property vectors as key and (ii) parametric Gaussian distribution as its corresponding value. We demonstrate the benefits of the learnt specific object-level conditioning distributions in multiple downstream tasks, namely object discovery, compositional scene generation, and compositional visual reasoning. We show that our method provides scene composition capabilities and a significant boost in a few shot adaptability tasks of compositional visual reasoning, while performing similarly or better than slot attention in object discovery tasks

Abstract: \emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \emph{label} given knowledge of the correct answer \emph{text}. We find that the existing techniques of logit attribution, attention pattern visualization, and activation patching naturally scale to Chinchilla, allowing us to identify and categorize a small set of `output nodes' (attention heads and MLPs). We further study the `correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results. For normal multiple-choice question answers, we significantly compress the query, key and value subspaces of the head without loss of performance when operating on the answer labels for multiple-choice questions, and we show that the query and key subspaces represent an `Nth item in an enumeration' feature to at least some extent. However, when we attempt to use this explanation to understand the heads' behaviour on a more general distribution including randomized answer labels, we find that it is only a partial explanation, suggesting there is more to learn about the operation of `correct letter' heads on multiple choice question answering.

Abstract: Modern language models can imitate complex patterns through few-shot learning, enabling them to complete challenging tasks without fine-tuning. However, imitation can also lead models to reproduce inaccuracies or harmful content if present in the context. We study harmful imitation through the lens of a model's internal representations, and identify two related phenomena: overthinking and false induction heads. The first phenomenon, overthinking, appears when we decode predictions from intermediate layers, given correct vs. incorrect few-shot demonstrations. At early layers, both demonstrations induce similar model behavior, but the behavior diverges sharply at some "critical layer", after which the accuracy given incorrect demonstrations progressively decreases. The second phenomenon, false induction heads, are a possible mechanistic cause of overthinking: these are heads in late layers that attend to and copy false information from previous demonstrations, and whose ablation reduces overthinking. Beyond scientific understanding, our results suggest that studying intermediate model computations could be a promising avenue for understanding and guarding against harmful model behaviors.

Abstract: Efficient and sustainable power generation is a crucial concern in the energy sector. In particular, thermal power plants grapple with accurately predicting steam mass flow, which is crucial for operational efficiency and cost reduction. In this study, we use a parallel hybrid neural network architecture that combines a parametrized quantum circuit and a conventional feed-forward neural network specifically designed for time-series prediction in industrial settings to enhance predictions of steam mass flow 15 minutes into the future. Our results show that the parallel hybrid model outperforms standalone classical and quantum models, achieving more than 5.7 and 4.9 times lower mean squared error (MSE) loss on the test set after training compared to pure classical and pure quantum networks, respectively. Furthermore, the hybrid model demonstrates smaller relative errors between the ground truth and the model predictions on the test set, up to 2 times better than the pure classical model. These findings contribute to the broader scientific understanding of how integrating quantum and classical machine learning techniques can be applied to real-world challenges faced by the energy sector, ultimately leading to optimized power plant operations.