XOR Codes and Sparse Learning Parity with Noise
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
A k-LIN instance is a system of m equations over n variables of the form si1 + · · · + sik = 0 or 1 modulo 2 (each involving k variables). We consider two distributions on instances in which the variables are chosen independently and uniformly but the right-hand sides are different. In a noisy planted instance, the right-hand side is obtained by evaluating the system on a random planted solution and adding independent noise with some constant bias to each equation; whereas in a random instance, the right-hand side is uniformly random. Alekhnovich (FOCS 2003) conjectured that the two are hard to distinguish when k = 3 and m = O(n). We give a sample-efficient reduction from solving noisy planted k-LIN instances (a sparse-equation version of the Learning Parity with Noise problem) to distinguishing them from random instances. Suppose that m-equation, n-variable instances of the two types are efficiently distinguishable with advantage ε. Then, we show that O(m · (m/ε)2/k)-equation, n-variable noisy planted k-LIN instances are efficiently solvable with probability exp –Õ((m/ε)6/k). Our solver has worse success probability but better sample complexity than Applebaum's (SICOMP 2013). We extend our techniques to show that this can generalize to (possibly non-linear) k-CSPs. The solver is based on a new approximate local list-decoding algorithm for the k-XOR code at large distances. The k-XOR encoding of a function F: ∑ → {–1, 1} is its k-th tensor power Fk(x1, …, xk) = F(x1) · · · F(xk). Given oracle access to a function G that µ-correlates with Fk, our algorithm, say for constant k, outputs the description of a message that Ω(µ1/k)-correlates with F with probability exp(–Õ(µ−4/k)). Previous decoders, for such k, have a worse dependence on µ (Levin, Combinatorica 1987) or do not apply to subconstant µ1/k. We also prove a new XOR lemma for this parameter regime. The decoder and its analysis rely on a new structure-versus-randomness dichotomy for general Boolean-valued functions over product sets, which may be of independent interest.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it