The paper constructs hash-linked evidence graphs that bind hardware measurement quantities to their verification records, enabling offline auditing with probabilistic matrix checks and security measures against probe attacks on GPUs.
hub
Silent data corruptions at scale
16 Pith papers cite this work. Polarity classification is still indexing.
16
Pith papers citing it
hub tools
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
A new fault-injection framework enables a systematic empirical study that produces 17 takeaways on error propagation in LLM inference and four software-only mitigation directions.
ITHICA generates functional tests via intra-thread instruction duplication and comparison, detecting 39% more defective servers than baseline methods on over 3000 real CPUs while revealing new defect behaviors.
FLARE uses pairwise coprime test vectors to create unique divisibility signatures that localize faulty rows in systolic arrays with one test pass and over 98% probability for 256x256 INT16 arrays.
Kernel Contracts is a specification language that formalizes correctness requirements for ML kernels to ensure consistent results across heterogeneous silicon platforms.
Boomslang introduces a front-end/back-end pipeline with superpositions in its IR to enable general-purpose checking of arbitrary transaction isolation levels via SMT solving.
DRIFT uses resilience analysis, targeted DVFS, and adaptive rollback ABFT to deliver 36% average energy savings or 1.7x speedup in diffusion model inference while preserving generation quality.
Bit flips in 3D Gaussian splatting are highly concentrated in effect with certain high-order bits corrupting up to 75.7% of the frame, but a support guard reduces the worst footprint to 11.68% while preserving clean performance and improving quality under accumulated faults.
Proposes zkVM-based protocol for verifiable frontier AI pre-training with committed specs, network observations, Merkle commitments, and FP precompiles, estimating 36-month POC at single-digit overhead.
Large-scale GPU fault injection shows NaN/inf outcomes are only 1% of SDC, single-bit flips under 40%, and corruption addresses are periodic, supporting distribution-aware modeling.
LLMs resist low-frequency permanent GPU faults but certain datapaths and precision formats trigger catastrophic training divergence even at moderate fault rates.
Gemini Ultra reaches human-expert performance on MMLU for the first time and sets new state-of-the-art results on 30 of 32 benchmarks, including all 20 multimodal ones tested.
MSET and CEP deliver higher reliability than SECDED ECC for CNNs and Vision Transformers with zero memory overhead and substantially lower area and delay.
An aging-aware adaptive voltage scaling framework for AI accelerators reduces predicted threshold voltage shifts by ~19% and aging degradation by up to 46% while saving 14% lifetime power by leveraging neural network resilience.
Kwai Keye-VL-2.0-30B-A3B is a 30B MoE model with 3B active parameters using DSA adaptation and MOPD distillation that reports SOTA results on video understanding and agent benchmarks.
AIReSim is a discrete event simulator for evaluating failure mitigation, recovery, and capacity planning decisions in large AI clusters.
citing papers explorer
-
Zero knowledge verification for frontier AI training is possible
Proposes zkVM-based protocol for verifiable frontier AI pre-training with committed specs, network observations, Merkle commitments, and FP precompiles, estimating 36-month POC at single-digit overhead.