Independence Tests for Language Models
- URL: http://arxiv.org/abs/2502.12292v1
- Date: Mon, 17 Feb 2025 20:01:08 GMT
- Title: Independence Tests for Language Models
- Authors: Sally Zhu, Ahmed Ahmed, Rohith Kuditipudi, Percy Liang,
- Abstract summary: Given the weights of two models, can we test whether they were trained independently?
We consider two settings: constrained and unconstrained.
We propose a new test which matches hidden activations between two models, and which is robust to adversarial transformations and to changes in model architecture.
- Score: 47.0749292650885
- License:
- Abstract: We consider the following problem: given the weights of two models, can we test whether they were trained independently -- i.e., from independent random initializations? We consider two settings: constrained and unconstrained. In the constrained setting, we make assumptions about model architecture and training and propose a family of statistical tests that yield exact p-values with respect to the null hypothesis that the models are trained from independent random initializations. These p-values are valid regardless of the composition of either model's training data; we compute them by simulating exchangeable copies of each model under our assumptions and comparing various similarity measures of weights and activations between the original two models versus these copies. We report the p-values from these tests on pairs of 21 open-weight models (210 total pairs) and correctly identify all pairs of non-independent models. Our tests remain effective even if one model was fine-tuned for many tokens. In the unconstrained setting, where we make no assumptions about training procedures, can change model architecture, and allow for adversarial evasion attacks, the previous tests no longer work. Instead, we propose a new test which matches hidden activations between two models, and which is robust to adversarial transformations and to changes in model architecture. The test can also do localized testing: identifying specific non-independent components of models. Though we no longer obtain exact p-values from this, empirically we find it behaves as one and reliably identifies non-independent models. Notably, we can use the test to identify specific parts of one model that are derived from another (e.g., how Llama 3.1-8B was pruned to initialize Llama 3.2-3B, or shared layers between Mistral-7B and StripedHyena-7B), and it is even robust to retraining individual layers of either model from scratch.
Related papers
- Model-free Methods for Event History Analysis and Efficient Adjustment (PhD Thesis) [55.2480439325792]
This thesis is a series of independent contributions to statistics unified by a model-free perspective.
The first chapter elaborates on how a model-free perspective can be used to formulate flexible methods that leverage prediction techniques from machine learning.
The second chapter studies the concept of local independence, which describes whether the evolution of one process is directly influenced by another.
arXiv Detail & Related papers (2025-02-11T19:24:09Z) - Model Provenance Testing for Large Language Models [14.949325775620439]
We develop a framework for testing model provenance: Whether one model is derived from another.
Our approach is based on the key observation that real-world model derivations preserve significant similarities in model outputs.
Using only black-box access to models, we employ multiple hypothesis testing to compare model similarities against a baseline established by unrelated models.
arXiv Detail & Related papers (2025-02-02T07:39:37Z) - Test-Time Alignment via Hypothesis Reweighting [56.71167047381817]
Large pretrained models often struggle with underspecified tasks.
We propose a novel framework to address the challenge of aligning models to test-time user intent.
arXiv Detail & Related papers (2024-12-11T23:02:26Z) - Self-Consistency of Large Language Models under Ambiguity [4.141513298907867]
This work presents an evaluation benchmark for self-consistency in cases of under-specification.
We conduct a series of behavioral experiments on the OpenAI model suite using an ambiguous integer sequence completion task.
We find that average consistency ranges from 67% to 82%, far higher than would be predicted if a model's consistency was random.
arXiv Detail & Related papers (2023-10-20T11:57:56Z) - Universal Semi-supervised Model Adaptation via Collaborative Consistency
Training [92.52892510093037]
We introduce a realistic and challenging domain adaptation problem called Universal Semi-supervised Model Adaptation (USMA)
We propose a collaborative consistency training framework that regularizes the prediction consistency between two models.
Experimental results demonstrate the effectiveness of our method on several benchmark datasets.
arXiv Detail & Related papers (2023-07-07T08:19:40Z) - A Simple Unified Approach to Testing High-Dimensional Conditional
Independences for Categorical and Ordinal Data [0.26651200086513094]
Conditional independence (CI) tests underlie many approaches to model testing and structure learning in causal inference.
Most existing CI tests for categorical and ordinal data stratify the sample by the conditioning variables, perform simple independence tests in each stratum, and combine the results.
Here we propose a simple unified CI test for ordinal and categorical data that maintains reasonable calibration and power in high dimensions.
arXiv Detail & Related papers (2022-06-09T08:56:12Z) - Uncertainty Estimation for Language Reward Models [5.33024001730262]
Language models can learn a range of capabilities from unsupervised training on text corpora.
It is often easier for humans to choose between options than to provide labeled data, and prior work has achieved state-of-the-art performance by training a reward model from such preference comparisons.
We seek to address these problems via uncertainty estimation, which can improve sample efficiency and robustness using active learning and risk-averse reinforcement learning.
arXiv Detail & Related papers (2022-03-14T20:13:21Z) - Conformal prediction for the design problem [72.14982816083297]
In many real-world deployments of machine learning, we use a prediction algorithm to choose what data to test next.
In such settings, there is a distinct type of distribution shift between the training and test data.
We introduce a method to quantify predictive uncertainty in such settings.
arXiv Detail & Related papers (2022-02-08T02:59:12Z) - ModelDiff: Testing-Based DNN Similarity Comparison for Model Reuse
Detection [9.106864924968251]
ModelDiff is a testing-based approach to deep learning model similarity comparison.
A study on mobile deep learning apps has shown the feasibility of ModelDiff on real-world models.
arXiv Detail & Related papers (2021-06-11T15:16:18Z) - How Can We Know When Language Models Know? On the Calibration of
Language Models for Question Answering [80.82194311274694]
We examine the question "how can we know when language models know, with confidence, the answer to a particular query?"
We examine three strong generative models -- T5, BART, and GPT-2 -- and study whether their probabilities on QA tasks are well calibrated.
We then examine methods to calibrate such models to make their confidence scores correlate better with the likelihood of correctness.
arXiv Detail & Related papers (2020-12-02T03:53:13Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.