Intelligence is a human construct to represent the ability to achieve goals. Given this wide berth, intelligence has been defined countless times, studied in a variety of ways and quantified using numerous measures. Understanding intelligence ultimately requires theory and quantification, both of which are elusive. My main objectives are to identify some of the central elements in and surrounding intelligence, discuss some of its challenges and propose a theory based on first principles. I focus on intelligence as defined by and for humans, frequently in comparison to machines, with the intention of setting the stage for more general characterizations in life, collectives, human designs such as AI and in non-designed physical and chemical systems. I discuss key features of intelligence, including path efficiency and goal accuracy, intelligence as a Black Box, environmental influences, flexibility to deal with surprisal, the regress of intelligence, the relativistic nature of intelligence and difficulty, and temporal changes in intelligence including its evolution. I present a framework for a first principles Theory of IntelligenceS (TIS), based on the quantifiable macro-scale system features of difficulty, surprisal and goal resolution accuracy. The proposed partitioning of uncertainty/solving and accuracy/understanding is particularly novel since it predicts that paths to a goal not only function to accurately achieve goals, but as experimentations leading to higher probabilities for future attainable goals and increased breadth to enter new goal spaces. TIS can therefore explain endeavors that do not necessarily affect Darwinian fitness, such as leisure, politics, games and art. I conclude with several conceptual advances of TIS including a compact mathematical form of surprisal and difficulty, the theoretical basis of TIS, and open questions.

Over recent years, an increasing amount of compute and data has been poured into training large language models (LLMs), usually by doing one-pass learning on as many tokens as possible randomly selected from large-scale web corpora. While training on ever-larger portions of the internet leads to consistent performance improvements, the size of these improvements diminishes with scale, and there has been little work exploring the effect of data selection on pre-training and downstream performance beyond simple de-duplication methods such as Min-Hash. Here, we show that careful data selection (on top of de-duplicated data) via pre-trained model embeddings can speed up training (20% efficiency gains) and improves average downstream accuracy on 16 NLP tasks (up to 2%) at the 6.7B model scale. Furthermore, we show that repeating data intelligently consistently outperforms baseline training (while repeating random data performs worse than baseline training). Our results indicate that clever data selection can significantly improve LLM pre-training, calls into question the common practice of training for a single epoch on as much data as possible, and demonstrates a path to keep improving our models past the limits of randomly sampling web data.

Critical learning periods are periods early in development where temporary sensory deficits can have a permanent effect on behavior and learned representations. Despite the radical differences between biological and artificial networks, critical learning periods have been empirically observed in both systems. This suggests that critical periods may be fundamental to learning and not an accident of biology. Yet, why exactly critical periods emerge in deep networks is still an open question, and in particular it is unclear whether the critical periods observed in both systems depend on particular architectural or optimization details. To isolate the key underlying factors, we focus on deep linear network models, and show that, surprisingly, such networks also display much of the behavior seen in biology and artificial networks, while being amenable to analytical treatment. We show that critical periods depend on the depth of the model and structure of the data distribution. We also show analytically and in simulations that the learning of features is tied to competition between sources. Finally, we extend our analysis to multi-task learning to show that pre-training on certain tasks can damage the transfer performance on new tasks, and show how this depends on the relationship between tasks and the duration of the pre-training stage. To the best of our knowledge, our work provides the first analytically tractable model that sheds light into why critical learning periods emerge in biological and artificial networks.

Curriculum learning (CL) posits that machine learning models -- similar to humans -- may learn more efficiently from data that match their current learning progress. However, CL methods are still poorly understood and, in particular for natural language processing (NLP), have achieved only limited success. In this paper, we explore why. Starting from an attempt to replicate and extend a number of recent curriculum methods, we find that their results are surprisingly brittle when applied to NLP. A deep dive into the (in)effectiveness of the curricula in some scenarios shows us why: when curricula are employed in combination with the popular Adam optimisation algorithm, they oftentimes learn to adapt to suboptimally chosen optimisation parameters for this algorithm. We present a number of different case studies with different common hand-crafted and automated CL approaches to illustrate this phenomenon, and we find that none of them outperforms optimisation with only Adam with well-chosen hyperparameters. As such, our results contribute to understanding why CL methods work, but at the same time urge caution when claiming positive results.

It is becoming increasingly important to understand the vulnerability of machine learning models to adversarial attacks. One of the fundamental problems in adversarial machine learning is to quantify how much training data is needed in the presence of evasion attacks, where data is corrupted at test time. In this thesis, we work with the exact-in-the-ball notion of robustness and study the feasibility of adversarially robust learning from the perspective of learning theory, considering sample complexity. We first explore the setting where the learner has access to random examples only, and show that distributional assumptions are essential. We then focus on learning problems with distributions on the input data that satisfy a Lipschitz condition and show that robustly learning monotone conjunctions has sample complexity at least exponential in the adversary's budget (the maximum number of bits it can perturb on each input). However, if the adversary is restricted to perturbing $O(\log n)$ bits, then one can robustly learn conjunctions and decision lists w.r.t. log-Lipschitz distributions. We then study learning models where the learner is given more power. We first consider local membership queries, where the learner can query the label of points near the training sample. We show that, under the uniform distribution, the exponential dependence on the adversary's budget to robustly learn conjunctions remains inevitable. We then introduce a local equivalence query oracle, which returns whether the hypothesis and target concept agree in a given region around a point in the training sample, and a counterexample if it exists. We show that if the query radius is equal to the adversary's budget, we can develop robust empirical risk minimization algorithms in the distribution-free setting. We give general query complexity upper and lower bounds, as well as for concrete concept classes.

Intertextual allusions hold a pivotal role in Classical Philology, with Latin authors frequently referencing Ancient Greek texts. Until now, the automatic identification of these intertextual references has been constrained to monolingual approaches, seeking parallels solely within Latin or Greek texts. In this study, we introduce SPhilBERTa, a trilingual Sentence-RoBERTa model tailored for Classical Philology, which excels at cross-lingual semantic comprehension and identification of identical sentences across Ancient Greek, Latin, and English. We generate new training data by automatically translating English texts into Ancient Greek. Further, we present a case study, demonstrating SPhilBERTa's capability to facilitate automated detection of intertextual parallels. Our models and resources are available at https://github.com/Heidelberg-NLP/ancient-language-models.

Computerized Adaptive Testing (CAT) is a widely used, efficient test mode that adapts to the examinee's proficiency level in the test domain. CAT requires pre-trained item profiles, for CAT iteratively assesses the student real-time based on the registered items' profiles, and selects the next item to administer using candidate items' profiles. However, obtaining such item profiles is a costly process that involves gathering a large, dense item-response data, then training a diagnostic model on the collected data. In this paper, we explore the possibility of leveraging response data collected in the CAT service. We first show that this poses a unique challenge due to the inherent selection bias introduced by CAT, i.e., more proficient students will receive harder questions. Indeed, when naively training the diagnostic model using CAT response data, we observe that item profiles deviate significantly from the ground-truth. To tackle the selection bias issue, we propose the user-wise aggregate influence function method. Our intuition is to filter out users whose response data is heavily biased in an aggregate manner, as judged by how much perturbation the added data will introduce during parameter estimation. This way, we may enhance the performance of CAT while introducing minimal bias to the item profiles. We provide extensive experiments to demonstrate the superiority of our proposed method based on the three public datasets and one dataset that contains real-world CAT response data.

User modeling, which learns to represent users into a low-dimensional representation space based on their past behaviors, got a surge of interest from the industry for providing personalized services to users. Previous efforts in user modeling mainly focus on learning a task-specific user representation that is designed for a single task. However, since learning task-specific user representations for every task is infeasible, recent studies introduce the concept of universal user representation, which is a more generalized representation of a user that is relevant to a variety of tasks. Despite their effectiveness, existing approaches for learning universal user representations are impractical in real-world applications due to the data requirement, catastrophic forgetting and the limited learning capability for continually added tasks. In this paper, we propose a novel continual user representation learning method, called TERACON, whose learning capability is not limited as the number of learned tasks increases while capturing the relationship between the tasks. The main idea is to introduce an embedding for each task, i.e., task embedding, which is utilized to generate task-specific soft masks that not only allow the entire model parameters to be updated until the end of training sequence, but also facilitate the relationship between the tasks to be captured. Moreover, we introduce a novel knowledge retention module with pseudo-labeling strategy that successfully alleviates the long-standing problem of continual learning, i.e., catastrophic forgetting. Extensive experiments on public and proprietary real-world datasets demonstrate the superiority and practicality of TERACON. Our code is available at https://github.com/Sein-Kim/TERACON.

According to US court papers and Sherborne School, Ms Turing, who legally changed her name from Julie Schwinghamer in 1988, removed the items without permission from archives given to the school in 1965 by the Turing family, in memory of the time he spent there as a pupil.

Regression analysis is a key area of interest in the field of data analysis and machine learning which is devoted to exploring the dependencies between variables, often using vectors. The emergence of high dimensional data in technologies such as neuroimaging, computer vision, climatology and social networks, has brought challenges to traditional data representation methods. Tensors, as high dimensional extensions of vectors, are considered as natural representations of high dimensional data. In this book, the authors provide a systematic study and analysis of tensor-based regression models and their applications in recent years. It groups and illustrates the existing tensor-based regression methods and covers the basics, core ideas, and theoretical characteristics of most tensor-based regression methods. In addition, readers can learn how to use existing tensor-based regression methods to solve specific regression tasks with multiway data, what datasets can be selected, and what software packages are available to start related work as soon as possible. Tensor Regression is the first thorough overview of the fundamentals, motivations, popular algorithms, strategies for efficient implementation, related applications, available datasets, and software resources for tensor-based regression analysis. It is essential reading for all students, researchers and practitioners of working on high dimensional data.