pith. sign in

arxiv: 2606.01168 · v1 · pith:Z55UXDCCnew · submitted 2026-05-31 · 💻 cs.CL

Thinking Economically: A Hierarchical Framework for Adaptive-Complexity Reasoning in LLMs

Yubo Gao , Haotian Wu , Hong Chen , Junquan Huang , Yibo Yan , Jungang Li , Zihao Dongfang , Sicheng Tao
show 3 more authors
Puay Siew Tan Jie Zhang Xuming Hu
This is my paper

Pith reviewed 2026-06-28 17:03 UTC · model grok-4.3

classification 💻 cs.CL
keywords adaptive reasoningchain-of-thought efficiencyhierarchical budgetingLLM overthinkingPareto optimizationperplexity signalsFisher Information pruningmath reasoning benchmarks
0
0 comments X

The pith

Hierarchical Adaptive Budgeter (HAB) lets LLMs allocate reasoning effort adaptively across problems and steps, outperforming uniform Chain-of-Thought in both accuracy and efficiency.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that Chain-of-Thought reasoning often wastes computation on overthinking because complexity differs across problems and within steps. It proposes HAB, which first decides how deep to reason for a given problem and then budgets tokens inside each step using signals from perplexity comparisons and an adaptive Pareto objective. A Fisher Information pruner adds training guidance to promote economical patterns. On GSM8K and MATH500, the resulting models produce shorter rationales that are also more accurate than standard CoT or other baselines.

Core claim

HAB is a training framework that predicts optimal reasoning depth for each problem at the inter-step level and learns step-specific token budgets at the intra-step level from PPL-derived comparisons together with an adaptive Pareto optimization objective, while a Fisher Information-based pruner supplies fine-grained guidance; this allows the generator to internalize economical reasoning and produces stronger accuracy-token trade-offs than standard CoT on GSM8K and MATH500.

What carries the argument

Hierarchical Adaptive Budgeter (HAB) with inter-step depth prediction and intra-step token budgeting driven by PPL comparisons, adaptive Pareto optimization, and Fisher Information pruning.

If this is right

  • Models trained this way surpass standard CoT in accuracy on math benchmarks while using fewer tokens.
  • The adaptive Pareto objective captures local quality-efficiency trade-offs at each step.
  • Fisher Information pruning provides training-time signals that encourage economical reasoning patterns.
  • Coarse-to-fine budgeting avoids the uniform compression used in prior efficiency methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Adaptive budgeting principles could extend to non-math reasoning tasks where overthinking also occurs.
  • Integrating such mechanisms might lower the overall computational cost of deploying reasoning LLMs at scale.
  • The two-granularity heterogeneity assumption suggests that single-level compression methods will remain suboptimal.

Load-bearing premise

Reasoning complexity is heterogeneous at two distinct granularities and PPL-derived step comparisons plus the adaptive Pareto objective can reliably capture the local quality-efficiency trade-off.

What would settle it

Running HAB on a dataset of problems with uniform reasoning complexity and finding no accuracy gain or token reduction relative to standard CoT.

Figures

Figures reproduced from arXiv: 2606.01168 by Haotian Wu, Hong Chen, Jie Zhang, Jungang Li, Junquan Huang, Puay Siew Tan, Sicheng Tao, Xuming Hu, Yibo Yan, Yubo Gao, Zihao Dongfang.

Figure 1
Figure 1. Figure 1: Comparison of performance of TokenSkip with different LLMs backbones under different retention ratios on the GSM8K dataset. To address this, several methods based on learned constraints have been proposed. These arXiv:2606.01168v1 [cs.CL] 31 May 2026 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Exploratory experiments on the necessity of dynamically allocating the number of reasoning steps tailored [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The overall framework of HAB. The Qwen-Max branch is used only during data preparation to construct [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Broader adaptability of HAB across datasets [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Chain-of-Thought (CoT) has significantly enhanced LLM reasoning, yet often incurs substantial computational overhead due to "overthinking": generating excessively long rationales without commensurate accuracy gains. Existing efficiency methods typically apply uniform compression, which overlooks a critical observation that reasoning complexity is heterogeneous at two distinct granularity: across different problems and within individual reasoning steps. This motivates our principle of Thinking Economically: intelligently allocating computational resources based on intrinsic task and step demands rather than pursuing uniform brevity. We propose Hierarchical Adaptive Budgeter (HAB), a training framework that operationalizes this principle through coarse-to-fine budgeting. At the inter-step level, HAB predicts the optimal reasoning depth for each problem. At the intra-step level, HAB learns step-specific token budgeting signals from PPL-derived step comparisons and an adaptive Pareto optimization objective that captures the local quality-efficiency trade-off, while a Fisher Information-based pruner further provides fine-grained training-time guidance, thereby encouraging the generator to internalize more economical reasoning patterns. Experiments on GSM8K and MATH500 show that HAB not only surpasses standard CoT in accuracy but also reduces token usage, achieving a stronger performance-efficiency trade-off than the compared baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces Hierarchical Adaptive Budgeter (HAB), a training framework for LLMs that implements 'Thinking Economically' by allocating compute via coarse-to-fine budgeting: inter-step prediction of optimal reasoning depth per problem, and intra-step token budgeting learned from PPL-derived step comparisons plus an adaptive Pareto optimization objective (with Fisher Information pruner for training guidance). Experiments on GSM8K and MATH500 claim that HAB exceeds standard CoT in accuracy while reducing token usage and achieving a superior accuracy-token trade-off versus baselines.

Significance. If the empirical gains are robust and the PPL-based intra-step signals are shown to track reasoning utility rather than mere predictability, the hierarchical budgeting principle could meaningfully advance efficiency methods beyond uniform compression. The work explicitly targets heterogeneity at both problem and step granularities, which is a plausible source of overthinking; reproducible code or ablations isolating the Pareto objective would strengthen its contribution.

major comments (2)
  1. [Abstract / intra-step budgeting description] Abstract and methods (intra-step component): the central attribution of accuracy gains plus token reduction to the hierarchical principle rests on PPL-derived step comparisons serving as a reliable proxy for local quality-efficiency trade-offs. Perplexity measures next-token predictability under the model distribution, not logical validity or contribution to final answer correctness; without explicit correlation analysis or ablations on GSM8K/MATH500 showing that lower-PPL steps improve downstream accuracy, the Pareto optimization may be optimizing the wrong objective, undermining the claim that HAB internalizes more economical patterns.
  2. [Abstract / adaptive Pareto optimization] Abstract (adaptive Pareto objective): parameters of the adaptive Pareto optimization are learned from the same data used to evaluate the final performance-efficiency trade-off. This creates a potential circularity that must be resolved by describing the exact optimization procedure, whether validation splits are held out, or how the objective avoids post-hoc selection; otherwise the reported stronger trade-off cannot be confidently attributed to the budgeting principle rather than fitting artifacts.
minor comments (2)
  1. [Abstract] The abstract states 'reasoning complexity is heterogeneous at two distinct granularity' (should be 'granularities').
  2. [Experiments] No error bars, number of runs, or statistical significance tests are mentioned for the accuracy and token-usage claims; these should be added to support the 'surpasses' and 'stronger trade-off' statements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our work. We address the two major comments point-by-point below and will revise the manuscript to provide the requested clarifications and analyses.

read point-by-point responses

  1. Referee: [Abstract / intra-step budgeting description] Abstract and methods (intra-step component): the central attribution of accuracy gains plus token reduction to the hierarchical principle rests on PPL-derived step comparisons serving as a reliable proxy for local quality-efficiency trade-offs. Perplexity measures next-token predictability under the model distribution, not logical validity or contribution to final answer correctness; without explicit correlation analysis or ablations on GSM8K/MATH500 showing that lower-PPL steps improve downstream accuracy, the Pareto optimization may be optimizing the wrong objective, undermining the claim that HAB internalizes more economical patterns.

    Authors: We acknowledge that perplexity fundamentally reflects next-token predictability rather than logical validity. In the current manuscript, the PPL-derived comparisons are used to identify lower-cost step variants that preserve answer correctness in aggregate, as evidenced by the reported accuracy gains alongside token reductions. However, we agree that an explicit correlation study between PPL signals and downstream accuracy would strengthen the justification. In the revision we will add such an analysis (including step-level accuracy correlations and ablations removing the PPL component) on both GSM8K and MATH500. revision: yes

  2. Referee: [Abstract / adaptive Pareto optimization] Abstract (adaptive Pareto objective): parameters of the adaptive Pareto optimization are learned from the same data used to evaluate the final performance-efficiency trade-off. This creates a potential circularity that must be resolved by describing the exact optimization procedure, whether validation splits are held out, or how the objective avoids post-hoc selection; otherwise the reported stronger trade-off cannot be confidently attributed to the budgeting principle rather than fitting artifacts.

    Authors: We agree that a clear description of the optimization procedure and data splits is required. The adaptive Pareto objective is optimized on a held-out validation portion of the training data, with final performance reported on the standard test splits of GSM8K and MATH500. We will expand the methods section to specify the exact training/validation split ratios, the optimization schedule, and confirmation that no test-set information influences the Pareto parameters. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper proposes HAB as a new training framework that uses PPL-derived step comparisons and an adaptive Pareto objective for intra-step token budgeting, alongside inter-step depth prediction and Fisher pruning. It then reports empirical results on GSM8K and MATH500 showing accuracy gains and token reductions versus CoT baselines. No load-bearing derivation step is exhibited that reduces by construction to its own inputs via equations, fitted parameters renamed as predictions, or self-citation chains. The central performance-efficiency claims rest on external experimental measurements rather than tautological redefinitions, so the derivation chain is self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no visibility into free parameters, axioms, or invented entities; the reported gains rest on unstated modeling choices inside the Pareto objective and the assumption that PPL comparisons are valid proxies for step quality.

pith-pipeline@v0.9.1-grok · 5770 in / 1146 out tokens · 20226 ms · 2026-06-28T17:03:32.823886+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

68 extracted references · 37 canonical work pages · 15 internal anchors

  1. [1]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

  2. [2]

    Advances in neural information processing systems , volume=

    Chain-of-thought prompting elicits reasoning in large language models , author=. Advances in neural information processing systems , volume=

  3. [3]

    The Llama 3 Herd of Models

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [4]

    Self-Consistency Improves Chain of Thought Reasoning in Language Models

    Self-consistency improves chain of thought reasoning in language models , author=. arXiv preprint arXiv:2203.11171 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [5]

    Advances in neural information processing systems , volume=

    Tree of thoughts: Deliberate problem solving with large language models , author=. Advances in neural information processing systems , volume=

  6. [6]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Graph of thoughts: Solving elaborate problems with large language models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  7. [7]

    arXiv preprint arXiv:2412.16964 , year=

    System-2 Mathematical Reasoning via Enriched Instruction Tuning , author=. arXiv preprint arXiv:2412.16964 , year=

    work page arXiv

  8. [8]

    Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs

    Do not think that much for 2+ 3=? on the overthinking of o1-like llms , author=. arXiv preprint arXiv:2412.21187 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models

    Stop overthinking: A survey on efficient reasoning for large language models , author=. arXiv preprint arXiv:2503.16419 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  10. [10]

    arXiv preprint arXiv:2501.12570 , year=

    O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning , author=. arXiv preprint arXiv:2501.12570 , year=

    work page arXiv

  11. [11]

    arXiv preprint arXiv:2503.10460 , year=

    Light-r1: Curriculum sft, dpo and rl for long cot from scratch and beyond , author=. arXiv preprint arXiv:2503.10460 , year=

    work page arXiv

  12. [12]

    arXiv preprint arXiv:2502.12067 , year=

    Tokenskip: Controllable chain-of-thought compression in llms , author=. arXiv preprint arXiv:2502.12067 , year=

    work page arXiv

  13. [13]

    CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation

    Codi: Compressing chain-of-thought into continuous space via self-distillation , author=. arXiv preprint arXiv:2502.21074 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    arXiv preprint arXiv:2502.12134 , year=

    Softcot: Soft chain-of-thought for efficient reasoning with llms , author=. arXiv preprint arXiv:2502.12134 , year=

    work page arXiv

  15. [15]

    Compressed Chain of Thought: Efficient Reasoning Through Dense Representations

    Compressed chain of thought: Efficient reasoning through dense representations , author=. arXiv preprint arXiv:2412.13171 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    Training Large Language Models to Reason in a Continuous Latent Space

    Training large language models to reason in a continuous latent space , author=. arXiv preprint arXiv:2412.06769 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Advances in neural information processing systems , volume=

    Large language models are zero-shot reasoners , author=. Advances in neural information processing systems , volume=

  18. [18]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Navigate through Enigmatic Labyrinth A Survey of Chain of Thought Reasoning: Advances, Frontiers and Future , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  19. [19]

    Findings of the Association for Computational Linguistics: ACL 2023 , pages=

    Towards Reasoning in Large Language Models: A Survey , author=. Findings of the Association for Computational Linguistics: ACL 2023 , pages=

    2023

  20. [20]

    Least-to-Most Prompting Enables Complex Reasoning in Large Language Models

    Least-to-most prompting enables complex reasoning in large language models , author=. arXiv preprint arXiv:2205.10625 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    arXiv preprint arXiv:2402.10400 , year=

    Chain of logic: Rule-based reasoning with large language models , author=. arXiv preprint arXiv:2402.10400 , year=

    work page arXiv

  22. [22]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Symbolic Chain-of-Thought Distillation: Small Models Can Also “Think” Step-by-Step , author=. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  23. [23]

    Findings of the Association for Computational Linguistics: ACL 2023 , pages=

    Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes , author=. Findings of the Association for Computational Linguistics: ACL 2023 , pages=

    2023

  24. [24]

    Advances in Neural Information Processing Systems , volume=

    Lst: Ladder side-tuning for parameter and memory efficient transfer learning , author=. Advances in Neural Information Processing Systems , volume=

  25. [25]

    arXiv preprint arXiv:2503.04472 , year=

    Dast: Difficulty-adaptive slow-thinking for large reasoning models , author=. arXiv preprint arXiv:2503.04472 , year=

    work page arXiv

  26. [26]

    Kimi k1.5: Scaling Reinforcement Learning with LLMs

    Kimi k1. 5: Scaling reinforcement learning with llms , author=. arXiv preprint arXiv:2501.12599 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  27. [27]

    Demystifying Long Chain-of-Thought Reasoning in LLMs

    Demystifying Long Chain-of-Thought Reasoning in LLMs , author=. arXiv preprint arXiv:2502.03373 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [28]

    arXiv preprint arXiv:2502.15589 , year=

    Lightthinker: Thinking step-by-step compression , author=. arXiv preprint arXiv:2502.15589 , year=

    work page arXiv

  29. [29]

    arXiv preprint arXiv:2503.01422 , year=

    Sampling-efficient test-time scaling: Self-estimating the best-of-n sampling in early decoding , author=. arXiv preprint arXiv:2503.01422 , year=

    work page arXiv

  30. [30]

    arXiv preprint arXiv:2411.01855 , year=

    Can language models learn to skip steps? , author=. arXiv preprint arXiv:2411.01855 , year=

    work page arXiv

  31. [31]

    Efficient Reasoning with Hidden Thinking

    Efficient Reasoning with Hidden Thinking , author=. arXiv preprint arXiv:2501.19201 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    arXiv preprint arXiv:2310.05393 , year=

    Hierarchical side-tuning for vision transformers , author=. arXiv preprint arXiv:2310.05393 , year=

    work page arXiv

  33. [33]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Quantized Side Tuning: Fast and Memory-Efficient Tuning of Quantized Large Language Models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  34. [34]

    IEEE Access , volume=

    Symbiotic Tuning: A Simple Approach for Enhancing Task Performance of Side-Tuning , author=. IEEE Access , volume=. 2025 , publisher=

    2025

  35. [35]

    Procedia Computer Science , volume=

    Ladder fine-tuning approach for sam integrating complementary network , author=. Procedia Computer Science , volume=. 2024 , publisher=

    2024

  36. [36]

    The eleventh international conference on learning representations , year=

    Automatic chain of thought prompting in large language models , author=. The eleventh international conference on learning representations , year=

  37. [37]

    Eric Zelikman and Yuhuai Wu and Jesse Mu and Noah Goodman , booktitle=

  38. [38]

    Advances in Neural Information Processing Systems , volume=

    Can Language Models Perform Robust Reasoning in Chain-of-thought Prompting with Noisy Rationales? , author=. Advances in Neural Information Processing Systems , volume=

  39. [39]

    Advances in Neural Information Processing Systems , volume=

    Iteration head: A mechanistic study of chain-of-thought , author=. Advances in Neural Information Processing Systems , volume=

  40. [40]

    arXiv preprint arXiv:2501.09804 , year=

    Enhancing Generalization in Chain of Thought Reasoning for Smaller Models , author=. arXiv preprint arXiv:2501.09804 , year=

    work page arXiv

  41. [41]

    Multimodal Chain-of-Thought Reasoning: A Comprehensive Survey

    Multimodal chain-of-thought reasoning: A comprehensive survey , author=. arXiv preprint arXiv:2503.12605 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  42. [42]

    The Thirteenth International Conference on Learning Representations , year=

    Unicott: A unified framework for structural chain-of-thought distillation , author=. The Thirteenth International Conference on Learning Representations , year=

  43. [43]

    Proceedings of the National Academy of Sciences , volume=

    Origins of the brain networks for advanced mathematics in expert mathematicians , author=. Proceedings of the National Academy of Sciences , volume=. 2016 , publisher=

    2016

  44. [44]

    NeuroImage , volume=

    A distinct cortical network for mathematical knowledge in the human brain , author=. NeuroImage , volume=. 2019 , publisher=

    2019

  45. [45]

    arXiv preprint arXiv:2411.19943 , year=

    Critical Tokens Matter: Token-Level Contrastive Estimation Enhence LLM's Reasoning Capability , author=. arXiv preprint arXiv:2411.19943 , year=

    work page arXiv

  46. [46]

    Training Verifiers to Solve Math Word Problems

    Training verifiers to solve math word problems , author=. arXiv preprint arXiv:2110.14168 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  47. [47]

    Measuring Mathematical Problem Solving With the MATH Dataset

    Measuring mathematical problem solving with the math dataset , author=. arXiv preprint arXiv:2103.03874 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  48. [48]

    arXiv preprint arXiv:2502.20122 , year=

    Self-training elicits concise reasoning in large language models , author=. arXiv preprint arXiv:2502.20122 , year=

    work page arXiv

  49. [49]

    From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step

    From explicit cot to implicit cot: Learning to internalize cot step by step , author=. arXiv preprint arXiv:2405.14838 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  50. [50]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    Token-budget-aware llm reasoning , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  51. [51]

    arXiv preprint arXiv:2505.15778 , year=

    Soft thinking: Unlocking the reasoning potential of llms in continuous concept space , author=. arXiv preprint arXiv:2505.15778 , year=

    work page arXiv

  52. [52]

    arXiv preprint arXiv:2502.18600 , year=

    Chain of draft: Thinking faster by writing less , author=. arXiv preprint arXiv:2502.18600 , year=

    work page arXiv

  53. [53]

    arXiv preprint arXiv:2503.01141 , year=

    How well do llms compress their own chain-of-thought? a token complexity approach , author=. arXiv preprint arXiv:2503.01141 , year=

    work page arXiv

  54. [54]

    2024 2nd International Conference on Foundation and Large Language Models (FLLM) , pages=

    The benefits of a concise chain of thought on problem-solving in large language models , author=. 2024 2nd International Conference on Foundation and Large Language Models (FLLM) , pages=. 2024 , organization=

    2024

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    Unlocking the capabilities of thought: A reasoning boundary framework to quantify and optimize chain-of-thought , author=. Advances in Neural Information Processing Systems , volume=

  56. [56]

    arXiv preprint arXiv:2307.15337 , year=

    Skeleton-of-thought: Prompting llms for efficient parallel generation , author=. arXiv preprint arXiv:2307.15337 , year=

    work page arXiv

  57. [57]

    arXiv preprint arXiv:2503.05179 , year=

    Sketch-of-thought: Efficient llm reasoning with adaptive cognitive-inspired sketching , author=. arXiv preprint arXiv:2503.05179 , year=

    work page arXiv

  58. [58]

    arXiv preprint arXiv:2502.13260 , year=

    Stepwise perplexity-guided refinement for efficient chain-of-thought reasoning in large language models , author=. arXiv preprint arXiv:2502.13260 , year=

    work page arXiv

  59. [59]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    C3ot: Generating shorter chain-of-thought without compromising effectiveness , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  60. [60]

    arXiv preprint arXiv:2010.02180 , year=

    Pareto probing: Trading off accuracy for complexity , author=. arXiv preprint arXiv:2010.02180 , year=

    work page arXiv 2010

  61. [61]

    Advances in neural information processing systems , volume=

    Pareto multi-task learning , author=. Advances in neural information processing systems , volume=

  62. [62]

    Transactions of the Association for Computational Linguistics , volume=

    Autopeft: Automatic configuration search for parameter-efficient fine-tuning , author=. Transactions of the Association for Computational Linguistics , volume=. 2024 , publisher=

    2024

  63. [63]

    International Conference on Machine Learning , pages=

    Group fisher pruning for practical network compression , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  64. [64]

    Quantum Machine Intelligence , volume=

    Adaptive pruning algorithm using a quantum Fisher information matrix for parameterized quantum circuits , author=. Quantum Machine Intelligence , volume=. 2024 , publisher=

    2024

  65. [65]

    Neural computation , volume=

    Mutual information, Fisher information, and population coding , author=. Neural computation , volume=. 1998 , publisher=

    1998

  66. [66]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. ICLR , volume=

  67. [67]

    Advances in neural information processing systems , volume=

    Multi-task learning as multi-objective optimization , author=. Advances in neural information processing systems , volume=

  68. [68]

    arXiv preprint arXiv:2007.08124 , year=

    Logiqa: A challenge dataset for machine reading comprehension with logical reasoning , author=. arXiv preprint arXiv:2007.08124 , year=

    work page arXiv 2007