Machine Learning (cs.LG)

Abstract: With various AI tools such as ChatGPT becoming increasingly popular, we are entering a true AI era. We can foresee that exceptional AI tools will soon reap considerable profits. A crucial question arise: should AI tools share revenue with their training data providers in additional to traditional stakeholders and shareholders? The answer is Yes. Large AI tools, such as large language models, always require more and better quality data to continuously improve, but current copyright laws limit their access to various types of data. Sharing revenue between AI tools and their data providers could transform the current hostile zero-sum game relationship between AI tools and a majority of copyrighted data owners into a collaborative and mutually beneficial one, which is necessary to facilitate the development of a virtuous cycle among AI tools, their users and data providers that drives forward AI technology and builds a healthy AI ecosystem. However, current revenue-sharing business models do not work for AI tools in the forthcoming AI era, since the most widely used metrics for website-based traffic and action, such as clicks, will be replaced by new metrics such as prompts and cost per prompt for generative AI tools. A completely new revenue-sharing business model, which must be almost independent of AI tools and be easily explained to data providers, needs to establish a prompt-based scoring system to measure data engagement of each data provider. This paper systematically discusses how to build such a scoring system for all data providers for AI tools based on classification and content similarity models, and outlines the requirements for AI tools or third parties to build it. Sharing revenue with data providers using such a scoring system would encourage more data owners to participate in the revenue-sharing program. This will be a utilitarian AI era where all parties benefit.

Abstract: Layer Normalization (LayerNorm) is an inherent component in all Transformer-based models. In this paper, we show that LayerNorm is crucial to the expressivity of the multi-head attention layer that follows it. This is in contrast to the common belief that LayerNorm's only role is to normalize the activations during the forward pass, and their gradients during the backward pass. We consider a geometric interpretation of LayerNorm and show that it consists of two components: (a) projection of the input vectors to a $d-1$ space that is orthogonal to the $\left[1,1,...,1\right]$ vector, and (b) scaling of all vectors to the same norm of $\sqrt{d}$. We show that each of these components is important for the attention layer that follows it in Transformers: (a) projection allows the attention mechanism to create an attention query that attends to all keys equally, offloading the need to learn this operation by the attention; and (b) scaling allows each key to potentially receive the highest attention, and prevents keys from being "un-select-able". We show empirically that Transformers do indeed benefit from these properties of LayeNorm in general language modeling and even in computing simple functions such as "majority". Our code is available at https://github.com/tech-srl/layer_norm_expressivity_role .

Abstract: Reinforcement learning (RL) agents are known to be vulnerable to evasion attacks during deployment. In single-agent environments, attackers can inject imperceptible perturbations on the policy or value network's inputs or outputs; in multi-agent environments, attackers can control an adversarial opponent to indirectly influence the victim's observation. Adversarial policies offer a promising solution to craft such attacks. Still, current approaches either require perfect or partial knowledge of the victim policy or suffer from sample inefficiency due to the sparsity of task-related rewards. To overcome these limitations, we propose the Intrinsically Motivated Adversarial Policy (IMAP) for efficient black-box evasion attacks in single- and multi-agent environments without any knowledge of the victim policy. IMAP uses four intrinsic objectives based on state coverage, policy coverage, risk, and policy divergence to encourage exploration and discover stronger attacking skills. We also design a novel Bias-Reduction (BR) method to boost IMAP further. Our experiments demonstrate the effectiveness of these intrinsic objectives and BR in improving adversarial policy learning in the black-box setting against multiple types of victim agents in various single- and multi-agent MuJoCo environments. Notably, our IMAP reduces the performance of the state-of-the-art robust WocaR-PPO agents by 34\%-54\% and achieves a SOTA attacking success rate of 83.91\% in the two-player zero-sum game YouShallNotPass.

Abstract: Bayesian optimization (BO) is a powerful tool for seeking the global optimum of black-box functions. While evaluations of the black-box functions can be highly costly, it is desirable to reduce the use of expensive labeled data. For the first time, we introduce a teacher-student model to exploit semi-supervised learning that can make use of large amounts of unlabelled data under the context of BO. Importantly, we show that the selection of the validation and unlabeled data is key to the performance of BO. To optimize the sampling of unlabeled data, we employ a black-box parameterized sampling distribution optimized as part of the employed bi-level optimization framework. Taking one step further, we demonstrate that the performance of BO can be further improved by selecting unlabeled data from a dynamically fitted extreme value distribution. Our BO method operates in a learned latent space with reduced dimensionality, making it scalable to high-dimensional problems. The proposed approach outperforms significantly the existing BO methods on several synthetic and real-world optimization tasks.

Abstract: We provide a psychometric-grounded exposition of bias and fairness as applied to a typical machine learning pipeline for affective computing. We expand on an interpersonal communication framework to elucidate how to identify sources of bias that may arise in the process of inferring human emotions and other psychological constructs from observed behavior. Various methods and metrics for measuring fairness and bias are discussed along with pertinent implications within the United States legal context. We illustrate how to measure some types of bias and fairness in a case study involving automatic personality and hireability inference from multimodal data collected in video interviews for mock job applications. We encourage affective computing researchers and practitioners to encapsulate bias and fairness in their research processes and products and to consider their role, agency, and responsibility in promoting equitable and just systems.

Abstract: In Causal Discovery with latent variables, We define two data paradigms: definite data: a single-skeleton structure with observed nodes single-value, and indefinite data: a set of multi-skeleton structures with observed nodes multi-value. Multi,skeletons induce low sample utilization and multi values induce incapability of the distribution assumption, both leading that recovering causal relations from indefinite data is, as of yet, largely unexplored. We design the causal strength variational model to settle down these two problems. Specifically, we leverage the causal strength instead of independent noise as latent variable to mediate evidence lower bound. By this design ethos, The causal strength of different skeletons is regarded as a distribution and can be expressed as a single-valued causal graph matrix. Moreover, considering the latent confounders, we disentangle the causal graph G into two relatisubgraphs O and C. O contains pure relations between observed nodes, while C represents the relations from latent variables to observed nodes. We summarize the above designs as Confounding Disentanglement Causal Discovery (biCD), which is tailored to learn causal representation from indefinite data under the latent confounding. Finally, we conduct comprehensive experiments on synthetic and real-world data to demonstrate the effectiveness of our method.

Abstract: There has been a rapid growth in biomedical literature, yet capturing the heterogeneity of the bibliographic information of these articles remains relatively understudied. Although graph mining research via heterogeneous graph neural networks has taken center stage, it remains unclear whether these approaches capture the heterogeneity of the PubMed database, a vast digital repository containing over 33 million articles. We introduce PubMed Graph Benchmark (PGB), a new benchmark dataset for evaluating heterogeneous graph embeddings for biomedical literature. PGB is one of the largest heterogeneous networks to date and consists of 30 million English articles. The benchmark contains rich metadata including abstract, authors, citations, MeSH terms, MeSH hierarchy, and some other information. The benchmark contains an evaluation task of 21 systematic reviews topics from 3 different datasets. In PGB, we aggregate the metadata associated with the biomedical articles from PubMed into a unified source and make the benchmark publicly available for any future works.

Abstract: Federated Learning (FL) enables training ML models on edge clients without sharing data. However, the federated model's performance on local data varies, disincentivising the participation of clients who benefit little from FL. Fair FL reduces accuracy disparity by focusing on clients with higher losses while personalisation locally fine-tunes the model. Personalisation provides a participation incentive when an FL model underperforms relative to one trained locally. For situations where the federated model provides a lower accuracy than a model trained entirely locally by a client, personalisation improves the accuracy of the pre-trained federated weights to be similar to or exceed those of the local client model. This paper evaluates two Fair FL (FFL) algorithms as starting points for personalisation. Our results show that FFL provides no benefit to relative performance in a language task and may double the number of underperforming clients for an image task. Instead, we propose Personalisation-aware Federated Learning (PaFL) as a paradigm that pre-emptively uses personalisation losses during training. Our technique shows a 50% reduction in the number of underperforming clients for the language task while lowering the number of underperforming clients in the image task instead of doubling it. Thus, evidence indicates that it may allow a broader set of devices to benefit from FL and represents a promising avenue for future experimentation and theoretical analysis.

Abstract: Generating explanations for reinforcement learning (RL) is challenging as actions may produce long-term effects on the future. In this paper, we develop a novel framework for explainable RL by learning a causal world model without prior knowledge of the causal structure of the environment. The model captures the influence of actions, allowing us to interpret the long-term effects of actions through causal chains, which present how actions influence environmental variables and finally lead to rewards. Different from most explanatory models which suffer from low accuracy, our model remains accurate while improving explainability, making it applicable in model-based learning. As a result, we demonstrate that our causal model can serve as the bridge between explainability and learning.

Abstract: Multi-domain learning (MDL) refers to simultaneously constructing a model or a set of models on datasets collected from different domains. Conventional approaches emphasize domain-shared information extraction and domain-private information preservation, following the shared-private framework (SP models), which offers significant advantages over single-domain learning. However, the limited availability of annotated data in each domain considerably hinders the effectiveness of conventional supervised MDL approaches in real-world applications. In this paper, we introduce a novel method called multi-domain contrastive learning (MDCL) to alleviate the impact of insufficient annotations by capturing both semantic and structural information from both labeled and unlabeled data.Specifically, MDCL comprises two modules: inter-domain semantic alignment and intra-domain contrast. The former aims to align annotated instances of the same semantic category from distinct domains within a shared hidden space, while the latter focuses on learning a cluster structure of unlabeled instances in a private hidden space for each domain. MDCL is readily compatible with many SP models, requiring no additional model parameters and allowing for end-to-end training. Experimental results across five textual and image multi-domain datasets demonstrate that MDCL brings noticeable improvement over various SP models.Furthermore, MDCL can further be employed in multi-domain active learning (MDAL) to achieve a superior initialization, eventually leading to better overall performance.

Abstract: Federated Learning (FL) aims to train machine learning models for multiple clients without sharing their own private data. Due to the heterogeneity of clients' local data distribution, recent studies explore the personalized FL that learns and deploys distinct local models with the help of auxiliary global models. However, the clients can be heterogeneous in terms of not only local data distribution, but also their computation and communication resources. The capacity and efficiency of personalized models are restricted by the lowest-resource clients, leading to sub-optimal performance and limited practicality of personalized FL. To overcome these challenges, we propose a novel approach named pFedGate for efficient personalized FL by adaptively and efficiently learning sparse local models. With a lightweight trainable gating layer, pFedGate enables clients to reach their full potential in model capacity by generating different sparse models accounting for both the heterogeneous data distributions and resource constraints. Meanwhile, the computation and communication efficiency are both improved thanks to the adaptability between the model sparsity and clients' resources. Further, we theoretically show that the proposed pFedGate has superior complexity with guaranteed convergence and generalization error. Extensive experiments show that pFedGate achieves superior global accuracy, individual accuracy and efficiency simultaneously over state-of-the-art methods. We also demonstrate that pFedGate performs better than competitors in the novel clients participation and partial clients participation scenarios, and can learn meaningful sparse local models adapted to different data distributions.

Abstract: A large-scale dynamic network (LDN) is a source of data in many big data-related applications due to their large number of entities and large-scale dynamic interactions. They can be modeled as a high-dimensional incomplete (HDI) tensor that contains a wealth of knowledge about time patterns. A Latent factorization of tensors (LFT) model efficiently extracts this time pattern, which can be established using stochastic gradient descent (SGD) solvers. However, LFT models based on SGD are often limited by training schemes and have poor tail convergence. To solve this problem, this paper proposes a novel nonlinear LFT model (MNNL) based on momentum-incorporated SGD, which extracts non-negative latent factors from HDI tensors to make training unconstrained and compatible with general training schemes, while improving convergence accuracy and speed. Empirical studies on two LDN datasets show that compared to existing models, the MNNL model has higher prediction accuracy and convergence speed.

Abstract: Recently, DeepNorm scales Transformers into extremely deep (i.e., 1000 layers) and reveals the promising potential of deep scaling. To stabilize the training of deep models, DeepNorm (Wang et al., 2022) attempts to constrain the model update to a constant value. Although applying such a constraint can benefit the early stage of model training, it may lead to undertrained models during the whole training procedure. In this paper, we propose BranchNorm, which dynamically rescales the non-residual branch of Transformer in accordance with the training period. BranchNorm not only theoretically stabilizes the training with smooth gradient norms at the early stage, but also encourages better convergence in the subsequent training stage. Experiment results on multiple translation tasks demonstrate that BranchNorm achieves a better trade-off between training stability and converge performance.

Abstract: Pseudo labeling is a popular and effective method to leverage the information of unlabeled data. Conventional instance-aware pseudo labeling methods often assign each unlabeled instance with a pseudo label based on its predicted probabilities. However, due to the unknown number of true labels, these methods cannot generalize well to semi-supervised multi-label learning (SSMLL) scenarios, since they would suffer from the risk of either introducing false positive labels or neglecting true positive ones. In this paper, we propose to solve the SSMLL problems by performing Class-distribution-Aware Pseudo labeling (CAP), which encourages the class distribution of pseudo labels to approximate the true one. Specifically, we design a regularized learning framework consisting of the class-aware thresholds to control the number of pseudo labels for each class. Given that the labeled and unlabeled examples are sampled according to the same distribution, we determine the thresholds by exploiting the empirical class distribution, which can be treated as a tight approximation to the true one. Theoretically, we show that the generalization performance of the proposed method is dependent on the pseudo labeling error, which can be significantly reduced by the CAP strategy. Extensive experimental results on multiple benchmark datasets validate that CAP can effectively solve the SSMLL problems.

Abstract: Over the last few years Explainable Clustering has gathered a lot of attention. Dasgupta et al. [ICML'20] initiated the study of explainable k-means and k-median clustering problems where the explanation is captured by a threshold decision tree which partitions the space at each node using axis parallel hyperplanes. Recently, Laber et al. [Pattern Recognition'23] made a case to consider the depth of the decision tree as an additional complexity measure of interest. In this work, we prove that even when the input points are in the Euclidean plane, then any depth reduction in the explanation incurs unbounded loss in the k-means and k-median cost. Formally, we show that there exists a data set X in the Euclidean plane, for which there is a decision tree of depth k-1 whose k-means/k-median cost matches the optimal clustering cost of X, but every decision tree of depth less than k-1 has unbounded cost w.r.t. the optimal cost of clustering. We extend our results to the k-center objective as well, albeit with weaker guarantees.

Abstract: When deploying artificial agents in real-world environments where they interact with humans, it is crucial that their behavior is aligned with the values, social norms or other requirements of that environment. However, many environments have implicit constraints that are difficult to specify and transfer to a learning agent. To address this challenge, we propose a novel method that utilizes the principle of maximum causal entropy to learn constraints and an optimal policy that adheres to these constraints, using demonstrations of agents that abide by the constraints. We prove convergence in a tabular setting and provide an approximation which scales to complex environments. We evaluate the effectiveness of the learned policy by assessing the reward received and the number of constraint violations, and we evaluate the learned cost function based on its transferability to other agents. Our method has been shown to outperform state-of-the-art approaches across a variety of tasks and environments, and it is able to handle problems with stochastic dynamics and a continuous state-action space.

Abstract: Graph Transformer is gaining increasing attention in the field of machine learning and has demonstrated state-of-the-art performance on benchmarks for graph representation learning. However, as current implementations of Graph Transformer primarily focus on learning representations of small-scale graphs, the quadratic complexity of the global self-attention mechanism presents a challenge for full-batch training when applied to larger graphs. Additionally, conventional sampling-based methods fail to capture necessary high-level contextual information, resulting in a significant loss of performance. In this paper, we introduce the Hierarchical Scalable Graph Transformer (HSGT) as a solution to these challenges. HSGT successfully scales the Transformer architecture to node representation learning tasks on large-scale graphs, while maintaining high performance. By utilizing graph hierarchies constructed through coarsening techniques, HSGT efficiently updates and stores multi-scale information in node embeddings at different levels. Together with sampling-based training methods, HSGT effectively captures and aggregates multi-level information on the hierarchical graph using only Transformer blocks. Empirical evaluations demonstrate that HSGT achieves state-of-the-art performance on large-scale benchmarks with graphs containing millions of nodes with high efficiency.

Abstract: Data augmentation is a widely used technique for improving model performance in machine learning, particularly in computer vision and natural language processing. Recently, there has been increasing interest in applying augmentation techniques to reinforcement learning (RL) problems, with a focus on image-based augmentation. In this paper, we explore a set of generic wrappers designed to augment RL environments with noise and encourage agent exploration and improve training data diversity which are applicable to a broad spectrum of RL algorithms and environments. Specifically, we concentrate on augmentations concerning states, rewards, and transition dynamics and introduce two novel augmentation techniques. In addition, we introduce a noise rate hyperparameter for control over the frequency of noise injection. We present experimental results on the impact of these wrappers on return using three popular RL algorithms, Soft Actor-Critic (SAC), Twin Delayed DDPG (TD3), and Proximal Policy Optimization (PPO), across five MuJoCo environments. To support the choice of augmentation technique in practice, we also present analysis that explores the performance these techniques across environments. Lastly, we publish the wrappers in our noisyenv repository for use with gym environments.

Abstract: Binary Neural Networks (BNNs) use 1-bit weights and activations to efficiently execute deep convolutional neural networks on edge devices. Nevertheless, the binarization of the first layer is conventionally excluded, as it leads to a large accuracy loss. The few works addressing the first layer binarization, typically increase the number of input channels to enhance data representation; such data expansion raises the amount of operations needed and it is feasible only on systems with enough computational resources. In this work, we present a new method to binarize the first layer using directly the 8-bit representation of input data; we exploit the standard bit-planes encoding to extract features bit-wise (using depth-wise convolutions); after a re-weighting stage, features are fused again. The resulting model is fully binarized and our first layer binarization approach is model independent. The concept is evaluated on three classification datasets (CIFAR10, SVHN and CIFAR100) for different model architectures (VGG and ResNet) and, the proposed technique outperforms state of the art methods both in accuracy and BMACs reduction.

Abstract: Federated learning is an important framework in modern machine learning that seeks to integrate the training of learning models from multiple users, each user having their own local data set, in a way that is sensitive to data privacy and to communication loss constraints. In clustered federated learning, one assumes an additional unknown group structure among users, and the goal is to train models that are useful for each group, rather than simply training a single global model for all users. In this paper, we propose a novel solution to the problem of clustered federated learning that is inspired by ideas in consensus-based optimization (CBO). Our new CBO-type method is based on a system of interacting particles that is oblivious to group memberships. Our model is motivated by rigorous mathematical reasoning, including a mean field analysis describing the large number of particles limit of our particle system, as well as convergence guarantees for the simultaneous global optimization of general non-convex objective functions (corresponding to the loss functions of each cluster of users) in the mean-field regime. Experimental results demonstrate the efficacy of our FedCBO algorithm compared to other state-of-the-art methods and help validate our methodological and theoretical work.

Abstract: Graph neural networks (GNNs) have achieved remarkable success in various real-world applications. However, recent studies highlight the vulnerability of GNNs to malicious perturbations. Previous adversaries primarily focus on graph modifications or node injections to existing graphs, yielding promising results but with notable limitations. Graph modification attack~(GMA) requires manipulation of the original graph, which is often impractical, while graph injection attack~(GIA) necessitates training a surrogate model in the black-box setting, leading to significant performance degradation due to divergence between the surrogate architecture and the actual victim model. Furthermore, most methods concentrate on a single attack goal and lack a generalizable adversary to develop distinct attack strategies for diverse goals, thus limiting precise control over victim model behavior in real-world scenarios. To address these issues, we present a gradient-free generalizable adversary that injects a single malicious node to manipulate the classification result of a target node in the black-box evasion setting. We propose Gradient-free Generalizable Single Node Injection Attack, namely G$^2$-SNIA, a reinforcement learning framework employing Proximal Policy Optimization. By directly querying the victim model, G$^2$-SNIA learns patterns from exploration to achieve diverse attack goals with extremely limited attack budgets. Through comprehensive experiments over three acknowledged benchmark datasets and four prominent GNNs in the most challenging and realistic scenario, we demonstrate the superior performance of our proposed G$^2$-SNIA over the existing state-of-the-art baselines. Moreover, by comparing G$^2$-SNIA with multiple white-box evasion baselines, we confirm its capacity to generate solutions comparable to those of the best adversaries.

Abstract: Obtaining high-quality data for collaborative training of machine learning models can be a challenging task due to A) the regulatory concerns and B) lack of incentive to participate. The first issue can be addressed through the use of privacy enhancing technologies (PET), one of the most frequently used one being differentially private (DP) training. The second challenge can be addressed by identifying which data points can be beneficial for model training and rewarding data owners for sharing this data. However, DP in deep learning typically adversely affects atypical (often informative) data samples, making it difficult to assess the usefulness of individual contributions. In this work we investigate how to leverage gradient information to identify training samples of interest in private training settings. We show that there exist techniques which are able to provide the clients with the tools for principled data selection even in strictest privacy settings.

Abstract: While off-policy reinforcement learning (RL) algorithms are sample efficient due to gradient-based updates and data reuse in the replay buffer, they struggle with convergence to local optima due to limited exploration. On the other hand, population-based algorithms offer a natural exploration strategy, but their heuristic black-box operators are inefficient. Recent algorithms have integrated these two methods, connecting them through a shared replay buffer. However, the effect of using diverse data from population optimization iterations on off-policy RL algorithms has not been thoroughly investigated. In this paper, we first analyze the use of off-policy RL algorithms in combination with population-based algorithms, showing that the use of population data could introduce an overlooked error and harm performance. To test this, we propose a uniform and scalable training design and conduct experiments on our tailored framework in robot locomotion tasks from the OpenAI gym. Our results substantiate that using population data in off-policy RL can cause instability during training and even degrade performance. To remedy this issue, we further propose a double replay buffer design that provides more on-policy data and show its effectiveness through experiments. Our results offer practical insights for training these hybrid methods.

Abstract: Many machine learning (ML) libraries are accessible online for ML practitioners. Typical ML pipelines are complex and consist of a series of steps, each of them invoking several ML libraries. In this demo paper, we present ExeKGLib, a Python library that allows users with coding skills and minimal ML knowledge to build ML pipelines. ExeKGLib relies on knowledge graphs to improve the transparency and reusability of the built ML workflows, and to ensure that they are executable. We demonstrate the usage of ExeKGLib and compare it with conventional ML code to show its benefits.

Abstract: We introduce Masked Trajectory Models (MTM) as a generic abstraction for sequential decision making. MTM takes a trajectory, such as a state-action sequence, and aims to reconstruct the trajectory conditioned on random subsets of the same trajectory. By training with a highly randomized masking pattern, MTM learns versatile networks that can take on different roles or capabilities, by simply choosing appropriate masks at inference time. For example, the same MTM network can be used as a forward dynamics model, inverse dynamics model, or even an offline RL agent. Through extensive experiments in several continuous control tasks, we show that the same MTM network -- i.e. same weights -- can match or outperform specialized networks trained for the aforementioned capabilities. Additionally, we find that state representations learned by MTM can significantly accelerate the learning speed of traditional RL algorithms. Finally, in offline RL benchmarks, we find that MTM is competitive with specialized offline RL algorithms, despite MTM being a generic self-supervised learning method without any explicit RL components. Code is available at https://github.com/facebookresearch/mtm

Abstract: Understanding the performance of machine learning (ML) models across diverse data distributions is critically important for reliable applications. Despite recent empirical studies positing a near-perfect linear correlation between in-distribution (ID) and out-of-distribution (OOD) accuracies, we empirically demonstrate that this correlation is more nuanced under subpopulation shifts. Through rigorous experimentation and analysis across a variety of datasets, models, and training epochs, we demonstrate that OOD performance often has a nonlinear correlation with ID performance in subpopulation shifts. Our findings, which contrast previous studies that have posited a linear correlation in model performance during distribution shifts, reveal a "moon shape" correlation (parabolic uptrend curve) between the test performance on the majority subpopulation and the minority subpopulation. This non-trivial nonlinear correlation holds across model architectures, hyperparameters, training durations, and the imbalance between subpopulations. Furthermore, we found that the nonlinearity of this "moon shape" is causally influenced by the degree of spurious correlations in the training data. Our controlled experiments show that stronger spurious correlation in the training data creates more nonlinear performance correlation. We provide complementary experimental and theoretical analyses for this phenomenon, and discuss its implications for ML reliability and fairness. Our work highlights the importance of understanding the nonlinear effects of model improvement on performance in different subpopulations, and has the potential to inform the development of more equitable and responsible machine learning models.

Abstract: Tabular data is one of the most commonly used types of data in machine learning. Despite recent advances in neural nets (NNs) for tabular data, there is still an active discussion on whether or not NNs generally outperform gradient-boosted decision trees (GBDTs) on tabular data, with several recent works arguing either that GBDTs consistently outperform NNs on tabular data, or vice versa. In this work, we take a step back and ask, 'does it matter?' We conduct the largest tabular data analysis to date, by comparing 19 algorithms across 176 datasets, and we find that the 'NN vs. GBDT' debate is overemphasized: for a surprisingly high number of datasets, either the performance difference between GBDTs and NNs is negligible, or light hyperparameter tuning on a GBDT is more important than selecting the best algorithm. Next, we analyze 965 metafeatures to determine what properties of a dataset make NNs or GBDTs better-suited to perform well. For example, we find that GBDTs are much better than NNs at handling skewed feature distributions, heavy-tailed feature distributions, and other forms of dataset irregularities. Our insights act as a guide for practitioners to decide whether or not they need to run a neural net to reach top performance on their dataset. Our codebase and all raw results are available at https://github.com/naszilla/tabzilla.

Abstract: Many contemporary generative models of molecules are variational auto-encoders of molecular graphs. One term in their training loss pertains to reconstructing the input, yet reconstruction capabilities of state-of-the-art models have not yet been thoroughly compared on a large and chemically diverse dataset. In this work, we show that when several state-of-the-art generative models are evaluated under the same conditions, their reconstruction accuracy is surprisingly low, worse than what was previously reported on seemingly harder datasets. However, we show that improving reconstruction does not directly lead to better sampling or optimization performance. Failed reconstructions from the MoLeR model are usually similar to the inputs, assembling the same motifs in a different way, and possess similar chemical properties such as solubility. Finally, we show that the input molecule and its failed reconstruction are usually mapped by the different encoders to statistically distinguishable posterior distributions, hinting that posterior collapse may not fully explain why VAEs are bad at reconstructing molecular graphs.

Abstract: Recent AI-assistant agents, such as ChatGPT, predominantly rely on supervised fine-tuning (SFT) with human annotations and reinforcement learning from human feedback (RLHF) to align the output of large language models (LLMs) with human intentions, ensuring they are helpful, ethical, and reliable. However, this dependence can significantly constrain the true potential of AI-assistant agents due to the high cost of obtaining human supervision and the related issues on quality, reliability, diversity, self-consistency, and undesirable biases. To address these challenges, we propose a novel approach called SELF-ALIGN, which combines principle-driven reasoning and the generative power of LLMs for the self-alignment of AI agents with minimal human supervision. Our approach encompasses four stages: first, we use an LLM to generate synthetic prompts, and a topic-guided method to augment the prompt diversity; second, we use a small set of human-written principles for AI models to follow, and guide the LLM through in-context learning from demonstrations (of principles application) to produce helpful, ethical, and reliable responses to user's queries; third, we fine-tune the original LLM with the high-quality self-aligned responses so that the resulting model can generate desirable responses for each query directly without the principle set and the demonstrations anymore; and finally, we offer a refinement step to address the issues of overly-brief or indirect responses. Applying SELF-ALIGN to the LLaMA-65b base language model, we develop an AI assistant named Dromedary. With fewer than 300 lines of human annotations (including < 200 seed prompts, 16 generic principles, and 5 exemplars for in-context learning). Dromedary significantly surpasses the performance of several state-of-the-art AI systems, including Text-Davinci-003 and Alpaca, on benchmark datasets with various settings.