Search

The current paradigm of evaluating Large Language Models (LLMs) through static benchmarks comes with significant limitations, such as vulnerability to data contamination and a lack of adaptability to the evolving capabilities of LLMs. Therefore, evaluation methods that can adapt and generate evaluation data with controlled complexity are urgently needed. In this work, we introduce Dynamic Evaluation of LLMs via Adaptive Reasoning Graph Evolvement (DARG) to dynamically extend current benchmarks with controlled complexity and diversity. Specifically, we first extract the reasoning graphs of data points in current benchmarks and then perturb the reasoning graphs to generate novel testing data. Such newly generated test samples can have different levels of complexity while maintaining linguistic diversity similar to the original benchmarks. We further use a code-augmented LLM to ensure the label correctness of newly generated data. We apply our DARG framework to diverse reasoning tasks in four domains with 15 state-of-the-art LLMs. Experimental results show that almost all LLMs experience a performance decrease with increased complexity and certain LLMs exhibit significant drops. Additionally, we find that LLMs exhibit more biases when being evaluated via the data generated by DARG with higher complexity levels. These observations provide useful insights into how to dynamically and adaptively evaluate LLMs. The code is available at https://github.com/SALT-NLP/DARG.

We revisit the following problem, proposed by Kolmogorov: given prescribed marginal distributions $F$ and $G$ for random variables $X,Y$ respectively, characterize the set of compatible distribution functions for the sum $Z=X+Y$. Bounds on the distribution function for $Z$ were given by Markarov (1982), and Frank et al. (1987), the latter using copula theory. However, though they obtain the same bounds, they make different assertions concerning their sharpness. In addition, their solutions leave some open problems in the case when the given marginal distribution functions are discontinuous. These issues have led to some confusion and erroneous statements in subsequent literature, which we correct. Kolmogorov's problem is closely related to inferring possible distributions for individual treatment effects $Y_1 - Y_0$ given the marginal distributions of $Y_1$ and $Y_0$; the latter being identified from a randomized experiment. We use our new insights to sharpen and correct results due to Fan and Park (2010) concerning individual treatment effects, and to fill some other logical gaps.

Certain pure-state symmetry-protected topological orders (SPT) can be used as a resource for transmitting quantum information. Here, we investigate the ability to transmit quantum information using decohered SPT states, and relate this property to the "strange correlation functions" which diagnose quantum many-body orders in these mixed-states. This perspective leads to the identification of a class of quantum channels -- termed symmetry-decoupling channels -- which do not necessarily preserve any weak or strong symmetries of the SPT state, but nevertheless protect quantum many-body order in the decohered mixed-state. We quantify the ability to transmit quantum information in decohered SPT states through the coherent quantum information, whose behavior is generally related to a decoding problem, whereby local measurements in the system are used to attempt to "learn" the symmetry charge of the SPT state before decoherence., Comment: 27 pages, 8 figures

To safely deploy deep learning models in the clinic, a quality assurance framework is needed for routine or continuous monitoring of input-domain shift and the models' performance without ground truth contours. In this work, cardiac substructure segmentation was used as an example task to establish a QA framework. A benchmark dataset consisting of Computed Tomography (CT) images along with manual cardiac delineations of 241 patients were collected, including one 'common' image domain and five 'uncommon' domains. Segmentation models were tested on the benchmark dataset for an initial evaluation of model capacity and limitations. An image domain shift detector was developed by utilizing a trained Denoising autoencoder (DAE) and two hand-engineered features. Another Variational Autoencoder (VAE) was also trained to estimate the shape quality of the auto-segmentation results. Using the extracted features from the image/segmentation pair as inputs, a regression model was trained to predict the per-patient segmentation accuracy, measured by Dice coefficient similarity (DSC). The framework was tested across 19 segmentation models to evaluate the generalizability of the entire framework. As results, the predicted DSC of regression models achieved a mean absolute error (MAE) ranging from 0.036 to 0.046 with an averaged MAE of 0.041. When tested on the benchmark dataset, the performances of all segmentation models were not significantly affected by scanning parameters: FOV, slice thickness and reconstructions kernels. For input images with Poisson noise, CNN-based segmentation models demonstrated a decreased DSC ranging from 0.07 to 0.41, while the transformer-based model was not significantly affected.

Large Language Models (LLMs) are capable of successfully performing many language processing tasks zero-shot (without training data). If zero-shot LLMs can also reliably classify and explain social phenomena like persuasiveness and political ideology, then LLMs could augment the Computational Social Science (CSS) pipeline in important ways. This work provides a road map for using LLMs as CSS tools. Towards this end, we contribute a set of prompting best practices and an extensive evaluation pipeline to measure the zero-shot performance of 13 language models on 25 representative English CSS benchmarks. On taxonomic labeling tasks (classification), LLMs fail to outperform the best fine-tuned models but still achieve fair levels of agreement with humans. On free-form coding tasks (generation), LLMs produce explanations that often exceed the quality of crowdworkers' gold references. We conclude that the performance of today's LLMs can augment the CSS research pipeline in two ways: (1) serving as zero-shot data annotators on human annotation teams, and (2) bootstrapping challenging creative generation tasks (e.g., explaining the underlying attributes of a text). In summary, LLMs are posed to meaningfully participate in social science analysis in partnership with humans., Comment: To appear in "Computational Linguistics" (CL)

We propose a general framework for using local measurements, local unitaries, and non-local classical communication to construct quantum channels which can efficiently prepare mixed states with long-range quantum order or quantum criticality. As an illustration, symmetry-protected topological (SPT) phases can be universally converted into mixed-states with long-range entanglement, which can undergo phase transitions with quantum critical correlations of local operators and a logarithmic scaling of the entanglement negativity, despite coexisting with volume-law entropy. Within the same framework, we present two applications using fermion occupation number measurement to convert (i) spinful free fermions in one dimension into a quantum-critical mixed state with enhanced algebraic correlations between spins and (ii) Chern insulators into a mixed state with critical quantum correlations in the bulk. The latter is an example where mixed-state quantum criticality can emerge from a gapped state of matter in constant depth using local quantum operations and non-local classical communication., Comment: 25 pages, 11 figures; updated to the published version

Inspired by the coupled-layer construction of the X-Cube model, we introduce the X-Cube Floquet code, a dynamical quantum error-correcting code where the number of encoded logical qubits grows with system size. The X-Cube Floquet code is defined on a three-dimensional lattice, built from intersecting two-dimensional layers in the $xy$, $yz$, and $xz$ directions, and consists of a periodic sequence of two-qubit measurements which couple the layers together. Within a single Floquet cycle, the codespace switches between that of the X-Cube fracton order and layers of entangled, two-dimensional toric codes. The encoded logical qubits' dynamics are analyzed, and we argue that the new code has a non-zero error threshold. We provide a new Hamiltonian realization of the X-Cube model and, more generally, explore the phase diagram related to the sequence of measurements that define the X-Cube Floquet code., Comment: Main Text (6 pages, 5 figures), Appendices (4 pages, 5 figures)