Train 'n Trade: Foundations of Parameter Markets

Organizations typically train large models individually. This is costly and time-consuming, particularly for large-scale foundation models. Such vertical production is known to be suboptimal. Inspired by this economic insight, we ask whether it is possible to leverage others' expertise by trading the constituent parts in models, i.e., sets of weights, as if they were market commodities. While recent advances in aligning and interpolating models suggest that doing so may be possible, a number of fundamental questions must be answered to create viable parameter markets. In this work, we address these basic questions, propose a framework containing the infrastructure necessary for market operations to take place, study strategies for exchanging parameters, and offer means for agents to monetize parameters. Excitingly, compared to agents who train siloed models from scratch, we show that it is possible to mutually gain by using the market, even in competitive settings. This suggests that the notion of parameter markets may be a useful paradigm for improving large-scale model training in the future.

Paper

References (44)

Scroll for more · 32 remaining

Similar papers

Peer review

Reviewer Mu3U6/10 · confidence 3/52023-07-05

Summary

This paper introduces a novel concept called *parameter markets*, which serves as a platform for exchanging parameters learned by machine learning models. In this framework, agents have the option to engage in parameter trading, to achieve (1) mutual benefits through collaboration or (2) monetary gains by simply selling these parameters. The key contributions of this paper involve the formulation and design of these parameter markets. This encompasses defining the value of parameters, establishing pricing mechanisms, and outlining how transactions can occur. The authors explore both competitive and collaborative scenarios and illustrate the advantages agents can obtain by opting in to trade through the marketplace.

Strengths

1. The idea to trade parameters is novel (to my best knowledge) and is quite different from data markets. The key benefit from parameter markets is that agents can train models for their own tasks and can yet benefit from training runs of each other. This is a very relevant problem to study. 2. The formulation is thorough and the empirical experiments are adequate to demonstrate the claims made in the paper 3. The paper is well written and easy to follow.

Weaknesses

1. This work is a bit existential - this framework works if *model privacy and alignment are assured*. 2. Theorem 4.1 relies on the broker having access to $\theta^*$ which is a strong assumption. 3. There is no discussion relating to incentives (i.e. agent misreporting their valuations to benefit from the trades).Perhaps this paper could benefit from adding a mechanism design angle to this framework.

Questions

1. Refer to Weaknesses. 2. It's unclear why Cobb-Douglas utility function is used.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

4 excellent

Contribution

4 excellent

Limitations

To my best knowledge, there isn't an immediate potential negative societal impact of their work. This paper, however, doesn't have an explicit weakness section (relating to their framework).

Reviewer Tv8J5/10 · confidence 3/52023-07-06

Summary

* This paper proposes an economic framework for trading parameters of prediction models. * Interaction is modeled as a brokered marketplace - Each agent $u$ trains a model characterized by parameters $\theta_u$ using gradient descent. At each time-step $t$: * Each agent performs a gradient descent step on the previous parameters $\theta_u^{t-1}$, to obtain $\dot{\theta}_u^t$. * After the GD step, agents relay their model parameters $\dot{\theta}_u^t$ to a trusted broker, which calculates the optimal linear interpolation between them, yielding the possibly-improved combination $\bar{\theta}_u$ for each agent. The broker indicates for each agent the gain-from-trade $\Delta_u^t$. * The agents participate in Nash bargaining to decide on the value exchange, set trading prices, and $\theta^t_u$ is decided. Each agent has a valuation function $v_u$ for their own parameters, and the other agent’s parameters. Utility function is assumed to be fully known by the broker. * Analysis: * Analysis is restricted to the two-agent setting $u\in\{A,B\}$, and a specific form for gain-from-trade $\Delta_u^t$ is assumed. It is assumed that each seller has complete knowledge of the valuation prior, such that Myerson’s revenue-maximizing pricing mechanism (virtual valuations) can be applied based on the information relayed from the broker ($\alpha$, $\beta$, $\Delta_a^t$ - By Theorem 4.1). * It is assumed that the broker has complete knowledge of the optimal model parameters in advance ($\theta^*$). Combining with the assumptions about the agents’ utility function, a closed-form expression is obtained for the monetary transfer after Nash bargaining. * In Section 4.1, two upper bounds on convergence rates are presented - One for a scenario where agent $B$ always trades (Theorem 4.2), and one for a scenario where both agents don’t trade (Proposition 4.3). The “always-trade” bound (Thm 4.2) is lower, and the authors claim that this explains why trading is sometimes beneficial. * Two sets of experiments are presented: * The first set of experiments is an evaluation of the data-sharing setting, without taking parameters into account (Section 5.1.1 - complete networks, Section 5.1.2 - subsets of networks), on real-world vision datasets (MNIST, CIFAR10, TinyImageNet). * Finally, the value of trading is evaluated using a synthetic model, and positive results are reported for both cooperative and competitive scenarios.

Strengths

* Problem is well-motivated. * Leverages recent deep insights about the structure of contemporary learning problems. * The breadth of work is substantial, and it contains a number of new definitions, theoretical analyses, and empirical evaluations. * Code is provided.

Weaknesses

* Some key assumptions are not well supported: * Valuation functions $v(\theta)$ are assumed to be fully known by the broker. For contemporary neural nets, the dimensionality of $\theta$ could be in the order of billions (or even trillions, such as in GPT4), and therefore the parameter valuation would be a function $v:\mathbb{R}^{10^9}\to\mathbb{R}$ - Which may be prohibitively expensive to represent, store and compute. * Broker is assumed to know the true model parameters $\theta^*$ in advance, which in practice may render the whole learning process unnecessary. In L214, it is claimed that knowing $\theta^*$ “is not necessary in practice”, but I did not find further support of this claim. * Very specific functional forms are assumed for gain-from-trade but are not sufficiently justified. * A common prior on agent valuation is assumed (L208), however it is not discussed whether having such prior is realistic and how it would be implemented. * Mathematical soundness concern regarding one of the claims: In section 4.1, authors claim that trading is beneficial by comparing two performance upper bounds (Theorem 4.2, Proposition 4.3). However, the difference between two upper bounds does not indicate the relation between the true quantities unless they are tight (in other words, proving that $a<100$ and $b<50$ does not indicate that $a>b$ unless there are matching lower bounds for $a$ and $b$). Therefore, if this observation is true, the analysis in Section 4.1 does not address the posed question (whether participation leads to better convergence). * First set of experiments (Sec 5.1.1, 5.12) only seem to demonstrate the ability to interpolate model parameters, but do contain simulation of parameter trading. Economic behavior is only demonstrated on the synthetic dataset. As the benefit of parameter interpolation has already been demonstrated in the literature to some extent, I did not understand the contribution of the results in sections 5.1.1, 5.1.2 to the understanding of parameter-trading markets. * Paper only analyzes the case of two agents $\{A,B\}$. Unclear how results extend to multiple agents. * Trading parameters during the gradient descent process would require agents to train models simultaneously, or suffer significant delays due to trading. This may make such trading impractical.

Questions

* How would the model extend to more than 2 agents? Which results can be used as-is, and which need to be generalized? Would the system behave qualitatively differently in any way when the number of agents is increased? * What are the computational requirements from the broker? * Is it possible to quantify the economic implications of the broker not being able to interpolate successfully?

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

2 fair

Presentation

2 fair

Contribution

2 fair

Limitations

I feel that limitations could be discussed at more depth. The key assumptions are added gradually throughout the paper, and it is hard to keep track. I feel that the paper can greatly benefit from a thorough discussion of limitations.

Reviewer iN1r4/10 · confidence 2/52023-07-06

Summary

The paper investigates how to design a marketplace for model parameters. The marketplace consists of agents training models for potentially different objectives and a trusted third-party broker. The broker receives the model parameters from each agent, assesses (and informs each agent of) the loss achieved by the merged model, and helps determine prices using Nash bargaining. The paper provides a theoretical and empirical analysis of this framework. In particular, they prove a bound on the gain from trade for linear models and they analyze the effectiveness of buying parameters in terms of improving training convergence. They empirically assess their framework on an image classification task. They validate that the merging method of the broker (Ainsworth et al., 2022) improves loss, and they evaluate the effectiveness of their pricing strategy.

Strengths

The idea of trading subsets of parameters in a marketplace seems novel and interesting. Parameter marketplaces seem to contrast with classical approaches of data marketplaces, where agents buy data from each other, or model marketplaces, where agents directly purchase models or model access.

Weaknesses

The proposed framework seems to rely on there being a trusted broker facilitating the trades and assessing and reporting the gains of merging parameters to the agents. In reality, this seems to be a strong assumption, as it is not clear if (a) a broker can be trusted with all of the parameters, and (b) if a broker will be able to assess the gain of merging models without significant data and infrastructure. See question below. The technical contribution of the paper seems limited. The paper combines existing work on merging and aligning models (e.g. Ainsworth et al., 2022) and classical approaches for Nash bargaining. However, the paper seems to use these approaches out-of-box and does not seem to present significant new technical contributions. The paper is a bit difficult to follow and the exposition could be improved. For example, the results in Section 4 use a lot of notation and the qualitative insights are a bit difficult to infer.

Questions

How might the framework be modified to avoid the reliance on a trusted broker, e.g., via a more decentralized approach?

Rating

4: Borderline reject: Technically solid paper where reasons to reject, e.g., limited evaluation, outweigh reasons to accept, e.g., good evaluation. Please use sparingly.

Confidence

2: You are willing to defend your assessment, but it is quite likely that you did not understand the central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

2 fair

Presentation

2 fair

Contribution

2 fair

Limitations

The authors adequately addressed the limitations.

Reviewer 91tn6/10 · confidence 3/52023-07-10

Summary

The paper proposes a framework for collaborative and competitive parameter trading among deep learning agents. The authors conduct experiments to validate the effectiveness of the proposed framework in improving the performance of the agents. The experiments show that even when the agents are training on different tasks, they can still benefit from trading parameters. The authors also validate the proposed pricing mechanism in a competitive scenario. The paper makes contributions in demonstrating the potential of parameter trading to improve the performance of deep learning agents and providing a framework for collaborative and competitive trading.

Strengths

- The paper proposes a novel framework for collaborative and competitive parameter trading - The experimental results are promising. - The experiments show that even when the agents are training on different tasks, they can still benefit from trading parameters.

Weaknesses

- The authors do not provide an analysis of the limitations of the proposed framework, which could help identify potential issues in real-world scenarios. - The experiments could benefit from a more extensive evaluation of the proposed framework's performance by comparing it with more baselines. - The paper could benefit from a more detailed explanation of the proposed pricing mechanism and its implementation. - The authors do not discuss the potential ethical implications of parameter trading, which could be relevant in real-world applications.

Questions

- Why is parameter trading necessary? - How to resolve the legal concerns on copyrights?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

More limitations should be discussed.

Authorsrebuttal2023-08-11

Thank all reviewers again and keep further discussion

Dear Area Chair, We would love it if you could remind reviewers to kick off conversations with us! We believe we could easily answer any remaining questions and make any further clarifications. Additional feedback from reviewers would be invaluable. We would particularly love to engage with reviewers iN1r and Tv8J during the discussion period. Thank you so much for your help! The Authors

Reviewer iN1r2023-08-16

Thanks to the authors for responding to my questions. I appreciate the response and its discussion of the trusted broker assumption and the broker infrastructure. I have raised my score from a 3 (reject) to a 4 (borderline reject).

Authorsrebuttal2023-08-19

Thank you for raising your rating---we appreciate your engagement with our rebuttal! We have included new experimental results and added a discussion section on the study of brokers, trust, incentives, and infrastructure in our updated draft. We would like to ask whether there are any further questions we can answer. Thanks you!

Reviewer Mu3U2023-08-16

Thanks for the response!

Authorsrebuttal2023-08-19

Thank you for engaging with our rebuttal! We are excited to incorporate the finding regarding agent's incentives and clarity of utility function into the paper. We are eager to provide any further clarifications or address any additional questions. Thank you!

Reviewer Tv8J2023-08-18

Thank you for the detailed response. I appreciate the discussion, and the additional empirical results. I’m increasing my rating to 5.

Authorsrebuttal2023-08-19

Thank you for increasing your score and engaging with our rebuttal. We are excited to include the new experimental results in the updated version of our paper. We would love to answer any additional questions! Thank you!

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC