Pith. sign in

REVIEW 10 cited by

Characterising Bias in Compressed Models

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2010.03058 v2 pith:6IF4IUBB submitted 2020-10-06 cs.LG cs.AI

Sara Hooker , Nyalleng Moorosi , Gregory Clark , Samy Bengio , Emily Denton This is my paper
classification cs.LGcs.AI
keywords subsetcompressionexamplesaccuracyauditingbiasdisproportionatelyfurther
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

The popularity and widespread use of pruning and quantization is driven by the severe resource constraints of deploying deep neural networks to environments with strict latency, memory and energy requirements. These techniques achieve high levels of compression with negligible impact on top-line metrics (top-1 and top-5 accuracy). However, overall accuracy hides disproportionately high errors on a small subset of examples; we call this subset Compression Identified Exemplars (CIE). We further establish that for CIE examples, compression amplifies existing algorithmic bias. Pruning disproportionately impacts performance on underrepresented features, which often coincides with considerations of fairness. Given that CIE is a relatively small subset but a great contributor of error in the model, we propose its use as a human-in-the-loop auditing tool to surface a tractable subset of the dataset for further inspection or annotation by a domain expert. We provide qualitative and quantitative support that CIE surfaces the most challenging examples in the data distribution for human-in-the-loop auditing.

Discussion (0). Sign in to comment.

Forward citations

Cited by 10 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Wake Vision: A Tailored Dataset and Benchmark Suite for TinyML Computer Vision Applications

    cs.CV 2024-05 unverdicted novelty 7.0 of 10

    Wake Vision pipeline produces a 6M-image person detection dataset for TinyML with 2.2% label error, improving model accuracy up to 6.6% over prior VWW benchmark across architectures and subsets.

  2. QuantiBias: Benchmarking Quantization-Induced Bias in LLMs

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Quantization leaves refusal and multiple-choice bias checks flat while open-ended stereotype endorsement remains high (~24–27% under an independent judge), a gap standard safety evaluations miss.

  3. Don't Go Breaking My LLM: The Impact of Pruning Attention Layers on Explanation Faithfulness and Confidence Calibration

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    Pruning attention layers in five LLMs across eight datasets maintains accuracy but degrades faithfulness and calibration.

  4. On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Study

    cs.CL 2026-06 unverdicted novelty 6.0 of 10

    Systematic experiments reveal that activation steering trades fluency for concept control, is less effective on instruction-tuned models, and that prompting/SFT excel at injection but not removal, with textual metrics...

  5. Quantized Reasoning Models Think They Need to Think Longer, but They Do Not

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    Post-training quantization increases overthinking errors in reasoning models; a logit penalty on curated overthinking markers reduces CoT length 12-23% without accuracy loss.

  6. The Asymmetric Effects of Knowledge Distillation on Bias in Small Language Models

    cs.CL 2026-05 conditional novelty 6.0 of 10

    Response-based knowledge distillation makes small models less likely to pick stereotypes on unambiguous questions but destroys which specific ambiguous questions they abstain on, a per-item harm aggregate bias metrics hide.

  7. Weight Pruning Amplifies Bias: A Multi-Method Study of Compressed LLMs for Edge AI

    cs.LG 2026-05 conditional novelty 5.0 of 10

    Activation-aware pruning preserves perplexity but amplifies bias in LLMs, with 47-59% of previously neutral items developing new stereotypical responses at 70% sparsity.

  8. Bias In, Bias Out? Finding Unbiased Subnetworks in Vanilla Models

    cs.LG 2026-03 unverdicted novelty 5.0 of 10

    BISE extracts bias-free subnetworks from conventionally trained models via pruning, enabling debiased operation without retraining or additional data.

  9. The Uneven Impact of Post-Training Quantization in Machine Translation

    cs.CL 2025-08 conditional novelty 5.0 of 10

    Across five LLMs and four quantization methods, 4-bit compression mostly preserves translation quality for high-resource languages, while 2-bit compression disproportionately degrades low-resource and Indic languages,...

  10. Software Fairness: An Analysis and Survey

    cs.SE 2022-05 unverdicted novelty 4.0 of 10

    A literature survey of 164 papers on software fairness reveals gaps in requirements engineering, intersectional measures, unstructured data, and white-box ML methods.

Pith tools