pith. sign in

arxiv: 2607.00937 · v1 · pith:MNYS74RDnew · submitted 2026-07-01 · 💻 cs.CL

Persona Non Grata: LLM Persona-Driven Generations in MCQA are Unstable in Distinct Dimensions

C\'esar Guerra-Solano , Xiang Lorraine Li This is my paper

Pith reviewed 2026-07-02 13:04 UTC · model grok-4.3

classification 💻 cs.CL
keywords persona-driven generationsMCQAinstability metricsLLM consistencyprompt formatmodel familiesquestion domainstask accuracy
0
0 comments X

The pith

LLM persona-driven generations in multiple-choice question answering exhibit instability across performance, outcome, and question correctness that varies by model family, size, domain, and prompt format.

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

The paper develops three metrics to track distinct kinds of instability when large language models adopt personas while answering multiple-choice questions. Evaluation across models and tasks shows that instability is not uniform: it differs reliably between model families and sizes, rises for math and commonsense items, and responds more to task prompt format than to temperature. Because persona-driven generations are already common in research and applications, these patterns indicate that reliability cannot be assumed and that the most effective persona can change with small experimental choices.

Core claim

Persona-driven generations in MCQA are unstable in three distinct dimensions measured by performance stability, outcome stability, and question correctness stability. This instability varies consistently between model families and model sizes, is greater for math and commonsense questions, is more strongly affected by task prompt format than by other hyperparameters such as temperature, correlates with lower task accuracy, and produces different best and worst personas under different experimental settings.

What carries the argument

Three metrics—performance stability, outcome stability, and question correctness stability—that quantify separate dimensions of how persona assignments alter LLM predictions on multiple-choice items.

If this is right

  • Math and commonsense questions produce greater instability than other domains.
  • Task prompt format drives more prediction instability than temperature.
  • Higher instability is associated with lower overall task accuracy.
  • Different hyperparameter settings yield different rankings of best and worst personas for the same task.
  • Hyperparameter instability must be checked when selecting personas for MCQA.

Where Pith is reading between the lines

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

  • Users of persona methods in MCQA should test multiple prompt formats rather than relying on a single setting.
  • The stability metrics could be applied to non-MCQA tasks to check whether domain effects generalize.
  • Model choice for persona applications should weigh family-level and size-level consistency differences.

Load-bearing premise

The three metrics capture the relevant dimensions of instability without missing major failure modes or depending too heavily on arbitrary persona definitions and question selections.

What would settle it

A replication that finds prompt format does not increase instability more than temperature, or that math and commonsense questions do not show reliably higher instability than other domains under the same metrics, would falsify the central patterns.

Figures

Figures reproduced from arXiv: 2607.00937 by C\'esar Guerra-Solano, Xiang Lorraine Li.

Figure 1
Figure 1. Figure 1: Depicting instability in non-text-heavy persona-driven generations (PDGs). Given different, al￾though still reasonable, experiment configurations, rep￾resented by different settings for the task prompt format (TPF) hyperparameter, an LLM’s persona-driven perfor￾mance on a multiple-choice evaluation differs greatly. teacher. Explain [...]") or more specific task com￾pletion (e.g. "You are a hiring manager. … view at source ↗
Figure 2
Figure 2. Figure 2: A figure depicting how instability metrics are calculated based on PDGs for our dataset. Across runs in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A figure depicting the delta hyperparameter instability of temperature (Temp.), persona prompt format [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Depicting the delta hyperparameter instability [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Depicting the delta hyperparameter instability of Temp., PPF, and TPF compared to the baseline. "Dataset" [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: An example of different experimental out [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: A figure depicting the delta hyperparameter [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Instability results stratified by instability metric, by dataset category, by persona sub-category. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: All classifier performances when trained across 48 experimental settings, additionally stratified by specific [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparing instability results across different formulations of [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Hyperparameter instability results per metric, per model, stratified by dataset category and persona [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: All inter-persona performance distributions between the "unstable" and "stable" experiment settings, as [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
read the original abstract

Persona-driven generations (PDGs) have seen prolific use in research and industry applications, where a large language model (LLM) takes on a 'persona' while completing some task. While persona expressed through free-form text (like dialogue) has substantial work investigating stability or consistency, relatively, persona expressed in non-text-heavy outputs (like in multiple-choice question answering, or MCQA) is often overlooked. We work to address this gap, seeking to understand the instability of LLM PDGs in MCQA tasks. We develop three metrics investigating the performance, outcome, and question correctness stability, evaluating three distinct dimensions. Using these metrics, we find that instability varies consistently between model families and model size, and across question domains, with math/commonsense questions leading to greater instability. We also find task prompt format introduces more prediction instability than other hyperparameters, like temperature. Finally, we find that instability is related to task accuracy, and using our instability metrics, find different experimental settings that result in different best and worst personas for tasks, despite their similarity. This reveals the importance of checking hyperparameter instability in PDGs.

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

1 major / 2 minor

Summary. The manuscript empirically studies instability in LLM persona-driven generations (PDGs) for MCQA tasks. It defines three metrics capturing performance stability, outcome stability, and question-correctness stability, then reports that instability varies consistently by model family and size, is higher for math/commonsense domains than others, is more affected by task prompt format than by temperature, correlates with task accuracy, and produces different best/worst personas across experimental settings.

Significance. If the central claims hold after robustness verification, the work usefully distinguishes three dimensions of PDG instability and supplies concrete evidence that prompt format dominates other hyperparameters. This could inform more reliable use of personas in MCQA applications. The explicit comparison of instability across domains and the observation that optimal personas shift with settings are practical contributions, though the lack of checks on metric sensitivity to arbitrary persona and question choices limits immediate generalizability.

major comments (1)
  1. [§3 and §4] §3 (Metrics) and §4 (Experiments): the three stability metrics are defined and applied without any reported sensitivity analysis to persona phrasing, number of personas per category, or question sampling criteria. Because the claims of consistent patterns across model families, sizes, and domains (math/commonsense showing greater instability) rest on these metrics reflecting intrinsic properties rather than sampling artifacts, this is a load-bearing gap.
minor comments (2)
  1. [Abstract] Abstract: quantitative results, error bars, dataset sizes, and exclusion criteria are omitted, making it hard to gauge the strength of the reported patterns on first reading.
  2. [Experimental Setup] Ensure all model versions, exact prompt templates, and the full list of personas are provided in an appendix or repository to support reproducibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments on our work. The major concern identifies a gap in sensitivity analysis for the stability metrics. We address this point directly below and agree that additional verification will improve the manuscript.

read point-by-point responses

  1. Referee: [§3 and §4] §3 (Metrics) and §4 (Experiments): the three stability metrics are defined and applied without any reported sensitivity analysis to persona phrasing, number of personas per category, or question sampling criteria. Because the claims of consistent patterns across model families, sizes, and domains (math/commonsense showing greater instability) rest on these metrics reflecting intrinsic properties rather than sampling artifacts, this is a load-bearing gap.

    Authors: We agree that the manuscript does not report sensitivity analyses varying persona phrasing, the number of personas per category, or question sampling criteria. The experiments used fixed but diverse persona sets and question samples drawn from established benchmarks, and the observed patterns (e.g., higher instability in math/commonsense domains, prompt format effects) were consistent across multiple model families and sizes. This provides indirect support that the metrics capture systematic differences rather than artifacts. Nevertheless, we recognize that explicit checks would strengthen the claims against sampling concerns. In the revised version we will add a dedicated subsection to §4 that reports results from (i) rephrased persona prompts, (ii) different numbers of personas per category, and (iii) alternative question sampling procedures, confirming that the main patterns remain stable. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical metrics and observations are self-contained

full rationale

This is an empirical study that defines three stability metrics (performance, outcome, question-correctness) explicitly at the outset and reports observed patterns across model families, sizes, domains, and hyperparameters. No equations, derivations, or predictions are present that reduce results to fitted inputs by construction, and no self-citation chains or uniqueness theorems are invoked as load-bearing premises. The central claims rest on direct measurement rather than any self-referential reduction, satisfying the criteria for a non-circular empirical analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Central claim rests on empirical observations from LLM experiments; no free parameters, invented entities, or non-standard axioms are introduced.

axioms (1)
  • domain assumption LLMs respond to persona instructions in free-form text prompts
    Foundational setup for all persona-driven generation experiments.

pith-pipeline@v0.9.1-grok · 5729 in / 1098 out tokens · 24118 ms · 2026-07-02T13:04:16.989782+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

51 extracted references · 27 canonical work pages · 4 internal anchors

  1. [1]

    Measuring nominal scale agreement among many raters , volume =

    Fleiss, Joseph , year =. Measuring nominal scale agreement among many raters , volume =. Psychological Bulletin , doi =

  2. [2]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Personalized dialogue generation with persona-adaptive attention , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  3. [3]

    Proceedings of the Twelfth International Conference on Learning Representations (ICLR 2024) , year =

    What’s in my big data? , author =. Proceedings of the Twelfth International Conference on Learning Representations (ICLR 2024) , year =

    2024

  4. [4]

    Proceedings of the 2013 workshop on Automated knowledge base construction , pages=

    Reporting bias and knowledge acquisition , author=. Proceedings of the 2013 workshop on Automated knowledge base construction , pages=

    2013

  5. [5]

    and Haberland, Matt and Reddy, Tyler and Cournapeau, David and Burovski, Evgeni and Peterson, Pearu and Weckesser, Warren and Bright, Jonathan and

    Virtanen, Pauli and Gommers, Ralf and Oliphant, Travis E. and Haberland, Matt and Reddy, Tyler and Cournapeau, David and Burovski, Evgeni and Peterson, Pearu and Weckesser, Warren and Bright, Jonathan and. Nature Methods , year =

  6. [6]

    9th Python in Science Conference , year=

    statsmodels: Econometric and statistical modeling with python , author=. 9th Python in Science Conference , year=

  7. [7]

    P ea C o K : Persona Commonsense Knowledge for Consistent and Engaging Narratives

    Gao, Silin and Borges, Beatriz and Oh, Soyoung and Bayazit, Deniz and Kanno, Saya and Wakaki, Hiromi and Mitsufuji, Yuki and Bosselut, Antoine. P ea C o K : Persona Commonsense Knowledge for Consistent and Engaging Narratives. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023. doi:10.1865...

    work page doi:10.18653/v1/2023.acl-long.362 2023

  8. [8]

    Persona Expansion with Commonsense Knowledge for Diverse and Consistent Response Generation

    Kim, Donghyun and Ahn, Youbin and Kim, Wongyu and Lee, Chanhee and Lee, Kyungchan and Lee, Kyong-Ho and Kim, Jeonguk and Shin, Donghoon and Lee, Yeonsoo. Persona Expansion with Commonsense Knowledge for Diverse and Consistent Response Generation. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. 2...

    work page doi:10.18653/v1/2023.eacl-main.81 2023

  9. [9]

    2024 , publisher=

    Self-rag: Learning to retrieve, generate, and critique through self-reflection , author=. 2024 , publisher=

    2024

  10. [10]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Incorporating geo-diverse knowledge into prompting for increased geographical robustness in object recognition , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  11. [11]

    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=

  12. [12]

    arXiv preprint arXiv:2411.10541 , year=

    Does prompt formatting have any impact on llm performance? , author=. arXiv preprint arXiv:2411.10541 , year=

    work page arXiv

  13. [13]

    arXiv preprint arXiv:2410.02185 , year=

    Posix: A prompt sensitivity index for large language models , author=. arXiv preprint arXiv:2410.02185 , year=

    work page arXiv

  14. [14]

    arXiv preprint arXiv:2503.16527 , year=

    LLM Generated Persona is a Promise with a Catch , author=. arXiv preprint arXiv:2503.16527 , year=

    work page arXiv

  15. [15]

    arXiv preprint arXiv:2412.15291 , year=

    A large-scale simulation on large language models for decision-making in political science , author=. arXiv preprint arXiv:2412.15291 , year=

    work page arXiv

  16. [16]

    Can LLM be a Personalized Judge?

    Dong, Yijiang River and Hu, Tiancheng and Collier, Nigel. Can LLM be a Personalized Judge?. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.592

    work page doi:10.18653/v1/2024.findings-emnlp.592 2024

  17. [17]

    arXiv preprint arXiv:2305.18189 , year=

    Marked personas: Using natural language prompts to measure stereotypes in language models , author=. arXiv preprint arXiv:2305.18189 , year=

    work page arXiv

  18. [18]

    arXiv preprint arXiv:2508.19764 , year=

    Principled Personas: Defining and Measuring the Intended Effects of Persona Prompting on Task Performance , author=. arXiv preprint arXiv:2508.19764 , year=

    work page arXiv

  19. [19]

    Persona-judge: Personalized Alignment of Large Language Models via Token-level Self-judgment

    Zhang, Xiaotian and Chen, Ruizhe and Feng, Yang and Liu, Zuozhu. Persona-judge: Personalized Alignment of Large Language Models via Token-level Self-judgment. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.260

    work page doi:10.18653/v1/2025.findings-acl.260 2025

  20. [20]

    International Conference on Automated Machine Learning , pages=

    Cost-effective hyperparameter optimization for large language model generation inference , author=. International Conference on Automated Machine Learning , pages=. 2023 , organization=

    2023

  21. [21]

    arXiv preprint arXiv:2406.11107 , year=

    Exploring safety-utility trade-offs in personalized language models , author=. arXiv preprint arXiv:2406.11107 , year=

    work page arXiv

  22. [22]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    MockLLM: A Multi-Agent Behavior Collaboration Framework for Online Job Seeking and Recruiting , author=. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2 , pages=

  23. [23]

    arXiv preprint arXiv:2508.05622 , year=

    Simulating Human-Like Learning Dynamics with LLM-Empowered Agents , author=. arXiv preprint arXiv:2508.05622 , year=

    work page arXiv

  24. [24]

    Using LLM s to simulate students' responses to exam questions

    Benedetto, Luca and Aradelli, Giovanni and Donvito, Antonia and Lucchetti, Alberto and Cappelli, Andrea and Buttery, Paula. Using LLM s to simulate students' responses to exam questions. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.663

    work page doi:10.18653/v1/2024.findings-emnlp.663 2024

  25. [25]

    Enhancing responses from large language models with role-playing prompts: a comparative study on answering frequently asked questions about total knee arthroplasty

    Chen, Yi-Chen and Lee, Sheng-Hsun and Sheu, Huan and Lin, Sheng-Hsuan and Hu, Chih-Chien and Fu, Shih-Chen and Yang, Cheng-Pang and Lin, Yu-Chih. Enhancing responses from large language models with role-playing prompts: a comparative study on answering frequently asked questions about total knee arthroplasty. BMC Med. Inform. Decis. Mak

  26. [26]

    arXiv preprint arXiv:2505.17818 , year=

    PatientSim: A Persona-Driven Simulator for Realistic Doctor-Patient Interactions , author=. arXiv preprint arXiv:2505.17818 , year=

    work page arXiv

  27. [27]

    arXiv preprint arXiv:2404.13066 , year=

    Leveraging large language model as simulated patients for clinical education , author=. arXiv preprint arXiv:2404.13066 , year=

    work page arXiv

  28. [28]

    Are Personalized Stochastic Parrots More Dangerous? Evaluating Persona Biases in Dialogue Systems

    Wan, Yixin and Zhao, Jieyu and Chadha, Aman and Peng, Nanyun and Chang, Kai-Wei. Are Personalized Stochastic Parrots More Dangerous? Evaluating Persona Biases in Dialogue Systems. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023. doi:10.18653/v1/2023.findings-emnlp.648

    work page doi:10.18653/v1/2023.findings-emnlp.648 2023

  29. [29]

    Two Tales of Persona in LLM s: A Survey of Role-Playing and Personalization

    Tseng, Yu-Min and Huang, Yu-Chao and Hsiao, Teng-Yun and Chen, Wei-Lin and Huang, Chao-Wei and Meng, Yu and Chen, Yun-Nung. Two Tales of Persona in LLM s: A Survey of Role-Playing and Personalization. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.969

    work page doi:10.18653/v1/2024.findings-emnlp.969 2024

  30. [30]

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

    Evaluating Very Long-Term Conversational Memory of LLM Agents , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  31. [31]

    Findings of the Association for Computational Linguistics ACL 2024 , pages=

    Evaluating Large Language Model Biases in Persona-Steered Generation , author=. Findings of the Association for Computational Linguistics ACL 2024 , pages=

    2024

  32. [32]

    Character.AI , howpublished =

    Character.AI. Character.AI , howpublished =

  33. [33]

    arXiv preprint arXiv:2411.00027 , year=

    Personalization of large language models: A survey , author=. arXiv preprint arXiv:2411.00027 , year=

    work page arXiv

  34. [34]

    Advances in neural information processing systems , volume=

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

  35. [35]

    Advances in neural information processing systems , volume=

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

  36. [36]

    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

  37. [37]

    Qwen3 Technical Report

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  38. [38]

    N orm A d: A Framework for Measuring the Cultural Adaptability of Large Language Models

    Rao, Abhinav Sukumar and Yerukola, Akhila and Shah, Vishwa and Reinecke, Katharina and Sap, Maarten. N orm A d: A Framework for Measuring the Cultural Adaptability of Large Language Models. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long...

    work page doi:10.18653/v1/2025.naacl-long.120 2025

  39. [39]

    Social IQ a: Commonsense Reasoning about Social Interactions

    Sap, Maarten and Rashkin, Hannah and Chen, Derek and Le Bras, Ronan and Choi, Yejin. Social IQ a: Commonsense Reasoning about Social Interactions. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019. doi:10.18653/v1/D19-1454

    work page doi:10.18653/v1/d19-1454 2019

  40. [40]

    Measuring Massive Multitask Language Understanding

    Measuring massive multitask language understanding , author=. arXiv preprint arXiv:2009.03300 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  41. [41]

    Gupta, Shashank and Shrivastava, Vaishnavi and Deshpande, Ameet and Kalyan, Ashwin and Clark, Peter and Sabharwal, Ashish and Khot, Tushar , booktitle =. Bias

  42. [42]

    When ``A Helpful Assistant'' Is Not Really Helpful: Personas in System Prompts Do Not Improve Performances of Large Language Models

    Zheng, Mingqian and Pei, Jiaxin and Logeswaran, Lajanugen and Lee, Moontae and Jurgens, David. When ``A Helpful Assistant'' Is Not Really Helpful: Personas in System Prompts Do Not Improve Performances of Large Language Models. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.888

    work page doi:10.18653/v1/2024.findings-emnlp.888 2024

  43. [43]

    Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting

    Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting , author=. arXiv preprint arXiv:2310.11324 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    P ro SA : Assessing and Understanding the Prompt Sensitivity of LLM s

    Zhuo, Jingming and Zhang, Songyang and Fang, Xinyu and Duan, Haodong and Lin, Dahua and Chen, Kai. P ro SA : Assessing and Understanding the Prompt Sensitivity of LLM s. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.108

    work page doi:10.18653/v1/2024.findings-emnlp.108 2024

  45. [45]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  46. [46]

    Publications Manual , year = "1983", publisher =

    1983

  47. [47]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    work page doi:10.1145/322234.322243 1981

  48. [48]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  49. [49]

    Dan Gusfield , title =. 1997

    1997

  50. [50]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  51. [51]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =