Representation-Level Control Surfaces for Reliability Sensing

Implements the Representation-Level Control Surfaces (RLCS) paradigm for ensuring the reliability of autonomous systems and AI models. It provides three deterministic sensors: Residual Likelihood (ResLik) for population-level anomaly detection, Temporal Consistency Sensor (TCS) for drift and shock detection, and Agreement Sensor for multi-modal redundancy checks. These sensors feed into a standardized control surface that issues 'PROCEED', 'DEFER', or 'ABSTAIN' signals based on strict safety invariants, allowing systems to detect and react to out-of-distribution states, sensor failures, and environmental shifts before they propagate to decision-making layers.


resLIK

Deterministic reliability sensors operating on latent representations. This package implements the ResLik, TCS, and Agreement sensors along with a deterministic control surface.

Version: 0.1.1

Control Signal Semantics

The system issues standardized signals. These signals are semantic contracts, not errors.

Signal Meaning Recommended Action
PROCEED High confidence Safe to continue.
DEFER Uncertainty / Drift Not an error. Pause, inspect, log, or retry.
ABSTAIN Fundamental Invalidity Stop. Fallback to safety model.

Usage

ResLik Sensor

library(resLIK)
set.seed(42)
z <- matrix(rnorm(20), nrow=2)
out <- reslik(z, ref_mean=0, ref_sd=1)
print(out$gated)
print(out$diagnostics)

Temporal Consistency Sensor

z_t <- c(1, 0, 0)
z_prev <- c(0.9, 0.1, 0)
out <- tcs(z_t, z_prev)
print(out$consistency)

Agreement Sensor

z1 <- c(1, 0)
z2 <- c(0, 1)
out <- agreement(z1, z2)
print(out$agreement)

Control Surface

# Assuming 'res', 'tcs_out', 'ag_out' from above
decision <- rlcs_control(out, tcs=list(consistency=0.9), agreement=list(agreement=0.9))
print(decision)

Note on Folder Name

The package name is resLIK. The source folder is named resLIK_package to avoid conflicts on case-insensitive filesystems with the existing reslik directory.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("resLIK")

0.1.2 by MD. Arshad, 8 months ago


Browse source code at https://github.com/cran/resLIK


Authors: MD. Arshad [aut, cre]


Documentation:   PDF Manual  


MIT + file LICENSE license


Imports utils

Suggests testthat, stats, knitr, rmarkdown


See at CRAN