We present TuringQ, the first benchmark designed to evaluate the reasoning capabilities of large language models (LLMs) in the theory of computation. TuringQ consists of 4,006 undergraduate and graduate-level question-answer pairs, categorized into four difficulty levels and covering seven core theoretical areas. We evaluate several open-source LLMs, as well as GPT-4, using Chain of Thought prompting and expert human assessment. Additionally, we propose an automated LLM-based evaluation system that demonstrates competitive accuracy when compared to human evaluation. Fine-tuning a Llama3-8B model on TuringQ shows measurable improvements in reasoning ability and out-of-domain tasks such as algebra. TuringQ serves as both a benchmark and a resource for enhancing LLM performance in complex computational reasoning tasks. Our analysis offers insights into LLM capabilities and advances in AI comprehension of theoretical computer science.

Automatic short answer scoring (ASAS) helps reduce the grading burden on educators but often lacks detailed, explainable feedback. Existing methods in ASAS with feedback (ASAS-F) rely on fine-tuning language models with limited datasets, which is resource-intensive and struggles to generalize across contexts. Recent approaches using large language models (LLMs) have focused on scoring without extensive fine-tuning. However, they often rely heavily on prompt engineering and either fail to generate elaborated feedback or do not adequately evaluate it. In this paper, we propose a modular retrieval augmented generation based ASAS-F system that scores answers and generates feedback in strict zero-shot and few-shot learning scenarios. We design our system to be adaptable to various educational tasks without extensive prompt engineering using an automatic prompt generation framework. Results show an improvement in scoring accuracy by 9\% on unseen questions compared to fine-tuning, offering a scalable and cost-effective solution.

For tabular datasets, the change in the relationship between the label and covariates ($Y|X$-shifts) is common due to missing variables (a.k.a. confounders). Since it is impossible to generalize to a completely new and unknown domain, we study models that are easy to adapt to the target domain even with few labeled examples. We focus on building more informative representations of tabular data that can mitigate $Y|X$-shifts, and propose to leverage the prior world knowledge in LLMs by serializing (write down) the tabular data to encode it. We find LLM embeddings alone provide inconsistent improvements in robustness, but models trained on them can be well adapted/finetuned to the target domain even using 32 labeled observations. Our finding is based on a comprehensive and systematic study consisting of 7650 source-target pairs and benchmark against 261,000 model configurations trained by 22 algorithms. Our observation holds when ablating the size of accessible target data and different adaptation strategies. The code is available at https://github.com/namkoong-lab/LLM-Tabular-Shifts.

Efficiently learning a sequence of related tasks, such as in continual learning, poses a significant challenge for neural nets due to the delicate trade-off between catastrophic forgetting and loss of plasticity. We address this challenge with a grounded framework for sequentially learning related tasks based on Bayesian inference. Specifically, we treat the model's parameters as a nonlinear Gaussian state-space model and perform efficient inference using Gaussian filtering and smoothing. This general formalism subsumes existing continual learning approaches, while also offering a clearer conceptual understanding of its components. Leveraging Laplace approximations during filtering, we construct Gaussian posterior measures on the weight space of a neural network for each task. We use it as an efficient regularizer by exploiting the structure of the generalized Gauss-Newton matrix (GGN) to construct diagonal plus low-rank approximations. The dynamics model allows targeted control of the learning process and the incorporation of domain-specific knowledge, such as modeling the type of shift between tasks. Additionally, using Bayesian approximate smoothing can enhance the performance of task-specific models without needing to re-access any data.

Estimating the structure of directed acyclic graphs (DAGs, also known as Bayesian networks) is a challenging problem since the search space of DAGs is combinatorial and scales superexponentially with the number of nodes. Existing approaches rely on various local heuristics for enforcing the acyclicity constraint. In this paper, we introduce a fundamentally different strategy: we formulate the structure learning problem as a purely continuous optimization problem over real matrices that avoids this combinatorial constraint entirely. This is achieved by a novel characterization of acyclicity that is not only smooth but also exact. The resulting problem can be efficiently solved by standard numerical algorithms, which also makes implementation effortless.

We address the problem of algorithmic fairness: ensuring that sensitive information does not unfairly influence the outcome of a classifier. We present an approach based on empirical risk minimization, which incorporates a fairness constraint into the learning problem. It encourages the conditional risk of the learned classifier to be approximately constant with respect to the sensitive variable. We derive both risk and fairness bounds that support the statistical consistency of our methodology. We specify our approach to kernel methods and observe that the fairness requirement implies an orthogonality constraint which can be easily added to these methods. We further observe that for linear models the constraint translates into a simple data preprocessing step.

I'm taking over TechScape from Alex Hern, and I'd like to introduce myself and my ideas for this newsletter. A bit about me: I started working at the Guardian the day Sam Bankman-Fried went on trial. My first holiday from my new job coincided with the shock firing of Sam Altman from OpenAI. The story I tell over and over again at parties is the one about how I was arrested and jailed while reporting a story on deadly testicular injections. We'll dissect the significance of the week's most substantial tech news, investigate odd niches, catch you up on the best of the Guardian's reporting and offer a helpful tip now and then.

The labelling oracle will provide a noisy answer (with either adversarial noise or Tsybakov) In addition to the labeling oracle, there is a comparison oracle. The leaner then can ask which of two instances is more likely to be positive. This oracle's answers is also noisy. The authors provide results which show that if the learner uses comparison queries, then it can reduce the label-query complexity. The problem that the authors study seem to be a valid and interesting one.

The authors present an algorithm for transfer learning using a symbolic program synthesizer for finding the most adequate neural network architecture and selecting relevant neural network modules from previous tasks for transfer. The approach is heavily based on concepts from programming languages, but also studies the relevant concept of high-level transfer that is crucial for true lifelong learning. Results show how the algorithm is capable of selectively transferring (high- and low-level) knowledge in a meaningful way, and numerical results validate the significance of the approach. The authors claim that their method targets the lifelong learning problem, but theirs is really a transfer learning approach. Solving catastrophic forgetting by completely freezing the network parameters precludes the method from being true lifelong learning, in which the learning of subsequent tasks affects the performance of earlier tasks.

Summary: Meta-learning is motivated by the promise of being able to transfer knowledge from previous learning experiences to new task settings, such that a new task can be learned more effectively from few observations. Yet, updating highly parametric models with little amounts of data can easily lead to overfitting. A promising avenue towards overcoming this challenge is a Bayesian treatment of meta-learning. This work, builds on top of recent work that provides a Bayesian interpretation of MAML (model-agnostic-meta-learning). This contribution is a direct extension of (Grant et al 2018) - where the task-train posterior was approximated via a Gaussian distribution. Applying SVGD instead allows for a more flexible and (potentially) more accurate approximation of a highly complex posterior.