首页 > AI前沿 > From Constraints to Causality: Petri net Encodings for Explaining Binary Neural Network Inference

From Constraints to Causality: Petri net Encodings for Explaining Binary Neural Network Inference

arXiv机器学习 2026-11-17 16:43 7 阅读 查看原文
用户需要将一段纯文本的新闻正文排版成语义化的HTML片段。 规则要求: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 2. 每个自然段落用`

`包裹,长文用`

`/`

`划分小节。 3. 列表内容用`
  • `,引用用`
    `,代码用`
    `,重要句子可用``。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出``/``/``等文档级标签。
    
    正文内容是一段关于Binary Neural Networks (BNNs)的学术摘要/新闻。
    这段正文可以划分为几个逻辑部分:
    第一段:介绍BNN及其存在的问题。
    第二段:现有方法及其局限。
    第三段:本文提出的方法。
    第四段:实验验证与效果。
    第五段:未来方向。
    
    或者就当作一个自然段落,或者适当划分小节。
    考虑到正文不长,可以分为几个段落,或者加一个小标题。
    原文:
    "Binary Neural Networks (BNNs) are attractive for resource-constrained hardware and on-device components in autonomous systems. However, their discrete, non-linear behaviour makes them difficult to analyse and trust in safety-relevant settings. Existing approaches based on SAT and Satisfiability Modulo Theories can verify a BNN's input-output properties, but treat it as a monolithic constraint system, which reveal only whether a property holds and not how the computation reaches its result. In this paper, we present an early work that provides a complementary, discrete-event view. We give a compositional translation from the inference semantics of a fixed, trained BNN to a 1-safe Petri net (PN), where each computation step is an explicit transition, so that a firing sequence is a causal trace of the BNN's decision. On a worked example, we establish faithfulness via exhaustive input-output equivalence against a reference implementation, and show how contrasting firing sequences localises where a single-feature change alters the prediction. Finally, we outline two directions this opens: a correctness-of-translation theorem, and a causal-trace analysis of where two trained BNNs diverge in their decisions, a structural question that constraint-based verifiers cannot express."
    
    可以将其拆分为几个自然段落,或者添加一个小节标题。
    比如:
    

    研究背景与现有局限

    Binary Neural Networks (BNNs) are attractive for resource-constrained hardware and on-device components in autonomous systems. However, their discrete, non-linear behaviour makes them difficult to analyse and trust in safety-relevant settings. Existing approaches based on SAT and Satisfiability Modulo Theories can verify a BNN's input-output properties, but treat it as a monolithic constraint system, which reveal only whether a property holds and not how the computation reaches its result.

    本文方法与实验验证

    In this paper, we present an early work that provides a complementary, discrete-event view. We give a compositional translation from the inference semantics of a fixed, trained BNN to a 1-safe Petri net (PN), where each computation step is an explicit transition, so that a firing sequence is a causal trace of the BNN's decision. On a worked example, we establish faithfulness via exhaustive input-output equivalence against a reference implementation, and show how contrasting firing sequences localises where a single-feature change alters the prediction.

    未来研究方向

    Finally, we outline two directions this opens: a correctness-of-translation theorem, and a causal-trace analysis of where two trained BNNs diverge in their decisions, a structural question that constraint-based verifiers cannot express.

    检查规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。(满足) 2. 每个自然段落用

    包裹,长文用

    /

    划分小节。(满足,添加了

    划分小节,原文是一整段,拆分为三个小节段落,没有增删改内容,只是排版)。 等一下,规则4说“保留原文所有信息,不得增删改内容”。添加小标题算不算增删改内容?小标题是新增的内容。为了严格遵守“不得增删改内容”,我最好不要自己编造小标题文字。 那如果不加小标题,直接按句拆分或者就作为一个大段落? “长文用

    /

    划分小节”,如果原文没有小节标题,我直接把它分成几个

    即可,或者用

    但不用文字?不行。 或者我不添加虚构的小标题,直接将原文按逻辑分为几个

    段落。原文是一整段,我可以按句群拆分为几个

    ,不加

    /

    ,因为原文本身没有小节。 或者,我可以将第一句话作为引言,或者不使用

    ,只使用

    。 原文: "Binary Neural Networks (BNNs) are attractive for resource-constrained hardware and on-device components in autonomous systems. However, their discrete, non-linear behaviour makes them difficult to analyse and trust in safety-relevant settings. Existing approaches based on SAT and Satisfiability Modulo Theories can verify a BNN's input-output properties, but treat it as a monolithic constraint system, which reveal only whether a property holds and not how the computation reaches its result." ->

    "In this paper, we present an early work that provides a complementary, discrete-event view. We give a compositional translation from the inference semantics of a fixed, trained BNN to a 1-safe Petri net (PN), where each computation step is an explicit transition, so that a firing sequence is a causal trace of the BNN's decision." ->

    "On a worked example, we establish faithfulness via exhaustive input-output equivalence against a reference implementation, and show how contrasting firing sequences localises where a single-feature change alters the prediction." ->

    "Finally, we outline two directions this opens: a correctness-of-translation theorem, and a causal-trace analysis of where two trained BNNs diverge in their decisions, a structural question that constraint-based verifiers cannot express." ->

    重要句子可用,比如: In this paper, we present an early work that provides a complementary, discrete-event view. We give a compositional translation from the inference semantics of a fixed, trained BNN to a 1-safe Petri net (PN), where each computation step is an explicit transition, so that a firing sequence is a causal trace of the BNN's decision. 检查是否有代码、列表、引用?没有。所以只用

    和。 段落1:

    Binary Neural Networks (BNNs) are attractive for resource-constrained hardware and on-device components in autonomous systems. However, their discrete, non-linear behaviour makes them difficult to analyse and trust in safety-relevant settings. Existing approaches based on SAT and Satisfiability Modulo Theories can verify a BNN's input-output properties, but treat it as a monolithic constraint system, which reveal only whether a property holds and not how the computation reaches its result.

    段落2:

    In this paper, we present an early work that provides a complementary, discrete-event view. We give a compositional translation from the inference semantics of a fixed, trained BNN to a 1-safe Petri net (PN), where each computation step is an explicit transition, so that a firing sequence is a causal trace of the BNN's decision.

    段落3:

    On a worked example, we establish faithfulness via exhaustive input-output equivalence against a reference implementation, and show how contrasting firing sequences localises where a single-feature change alters the prediction.

    段落4:

    Finally, we outline two directions this opens: a correctness-of-translation theorem, and a causal-trace analysis of where two trained BNNs diverge in their decisions, a structural question that constraint-based verifiers cannot express.

    这样既满足了排版要求,又没有增删改任何内容(只是改变样式,不改变文本内容;拆分段落也是排版的合理操作,因为原文本就是由多个句号分隔的自然段逻辑,虽然原文