TCSLib

111 Learning Theory — Finite Minimax

111.1 Overview

This module develops finite two-player zero-sum games and proves an approximate minimax theorem via the Hedge (multiplicative weights) algorithm. The main results are weak duality—showing that the row player’s mixed-strategy minimum payoff never exceeds the column player’s mixed-strategy maximum payoff—and approx_minimax, which constructs an \(\varepsilon \)-approximate saddle point for any \(\varepsilon \gt 0\) by running Hedge for sufficiently many rounds against column best responses.

111.2 Declarations

Definition 111.1 Finite zero-sum game
#

A finite two-player zero-sum game with \(M\) row actions and \(N\) column actions is a record consisting of a payoff matrix \(A : \mathrm{Fin}\, M \to \mathrm{Fin}\, N \to \mathbb {R}\) together with proofs that every entry satisfies \(0 \le A_{ij} \le 1\). The row player seeks to maximise the payoff; the column player seeks to minimise it.

Definition 111.2 Mixed strategy
#

A mixed strategy over \(n\) actions is a record consisting of a weight function \(w : \mathrm{Fin}\, n \to \mathbb {R}\) together with proofs that every weight is non-negative and that the weights sum to \(1\), i.e. \(w\) is a probability distribution over \(\mathrm{Fin}\, n\).

Definition 111.3 Row mixed payoff against a pure column
#

Given a game \(G\), a mixed row strategy \(p\), and a pure column \(j\), the expected payoff against column \(j\) is

\[ \mathrm{payoffVsPure}(G, p, j) \; =\; \sum _{i} p_i \cdot A_{ij}. \]
Definition 111.4 Pure row payoff against a mixed column
#

Given a game \(G\), a pure row \(i\), and a mixed column strategy \(q\), the expected payoff of row \(i\) against \(q\) is

\[ \mathrm{pureVsPayoff}(G, i, q) \; =\; \sum _{j} A_{ij} \cdot q_j. \]
Definition 111.5 Column best response

Given a game \(G\) and a mixed row strategy \(p\), bestColumn is the pure column \(j^* \in \mathrm{Fin}\, N\) that minimises \(\mathrm{payoffVsPure}(G, p, \cdot )\), chosen via the classical axiom of choice from the finite minimiser.

Lemma 111.6 Best-column minimality

For any game \(G\), mixed row strategy \(p\), and pure column \(j\),

\[ \mathrm{payoffVsPure}(G, p, \mathrm{bestColumn}(G, p)) \; \le \; \mathrm{payoffVsPure}(G, p, j). \]

That is, the chosen column is indeed a minimiser.

Definition 111.7 Row best response
#

Given a game \(G\) and a mixed column strategy \(q\), bestRow is the pure row \(i^* \in \mathrm{Fin}\, M\) that maximises \(\mathrm{pureVsPayoff}(G, \cdot , q)\), chosen classically from the finite maximiser.

Lemma 111.8 Best-row maximality

For any game \(G\), mixed column strategy \(q\), and pure row \(i\),

\[ \mathrm{pureVsPayoff}(G, i, q) \; \le \; \mathrm{pureVsPayoff}(G, \mathrm{bestRow}(G, q), q). \]
Theorem 111.9 Weak duality

For any game \(G\) and any mixed strategies \(p\) (row) and \(q\) (column),

\[ \inf _{j \in \mathrm{Fin}\, N} \mathrm{payoffVsPure}(G, p, j) \; \le \; \sup _{i \in \mathrm{Fin}\, M} \mathrm{pureVsPayoff}(G, i, q). \]

The row player’s worst-case mixed payoff never exceeds the column player’s best-case mixed payoff.

Definition 111.10 Loss sequence induced by a game
#

Given a game \(G\) and a sequence of column responses \(j_0, \dots , j_{T-1}\), the induced loss sequence for the row player is \(\ell _t(i) = 1 - A_{i,j_t} \in [0,1]\). This translates a zero-sum game into the abstract loss-sequence framework used by Hedge.

Lemma 111.11 Induced loss sequence is valid

The loss sequence \(G.\mathrm{toLossSeq}(\mathrm{colResponse})\) is valid, i.e. every loss lies in \([0, 1]\), because the payoffs satisfy \(0 \le A_{ij} \le 1\).

Definition 111.12 Average strategy
#

Given \(T\) distributions \(p_0, \dots , p_{T-1}\) over \(\mathrm{Fin}\, n\) (each presented as non-negative weights summing to \(1\)) and \(T \gt 0\), the average strategy is the mixed strategy with weights

\[ \bar{p}_i \; =\; \frac{1}{T} \sum _{t=0}^{T-1} p_t(i). \]
Definition 111.13 Hedge mixed strategy

The Hedge distribution at round \(t\) for learning rate \(\eta \) and loss sequence \(\ell \), packaged as a MixedStrategy by bundling the non-negativity and sum-one proofs from hedgeDist.

Definition 111.14 Empirical strategy
#

Given \(T \gt 0\) pure actions \(a_0, \dots , a_{T-1} \in \mathrm{Fin}\, n\), the empirical strategy is the mixed strategy assigning to each \(i\) the fraction of rounds on which action \(i\) was played:

\[ q_i \; =\; \frac{1}{T}\# \{ t : a_t = i\} . \]
Lemma 111.15 Payoff under average strategy

For a game \(G\), a column \(j\), and \(T\) row distributions \(p_0, \dots , p_{T-1}\),

\[ \mathrm{payoffVsPure}(G, \bar{p}, j) \; =\; \frac{1}{T} \sum _{t=0}^{T-1} \sum _{i} p_t(i) \cdot A_{ij}, \]

where \(\bar{p}\) is the average strategy.

Lemma 111.16 Payoff of pure row under empirical column

For a game \(G\), a pure row \(i\), and a sequence of pure column actions \(a_0, \dots , a_{T-1}\),

\[ \mathrm{pureVsPayoff}(G, i, \hat{q}) \; =\; \frac{1}{T} \sum _{t=0}^{T-1} A_{i, a_t}, \]

where \(\hat{q}\) is the empirical column strategy.

Definition 111.17 Prefix cumulative game loss
#

Given a game \(G\), a prefix of \(t\) column actions, and a row \(i\), the prefix cumulative loss of row \(i\) is

\[ L_t(i) \; =\; \sum _{s=0}^{t-1} \bigl(1 - A_{i, a_s}\bigr). \]

This mirrors cumLoss from the abstract Hedge framework, expressed directly in terms of game payoffs.

Definition 111.18 Prefix Hedge weight
#

The unnormalised Hedge weight of row \(i\) after seeing a prefix of column actions at learning rate \(\eta \) is

\[ w_t(i) \; =\; \exp \! \bigl(-\eta \cdot L_t(i)\bigr), \]

where \(L_t(i)\) is the prefix cumulative loss.

Definition 111.19 Prefix potential
#

The prefix potential is the sum of prefix Hedge weights over all rows:

\[ \Phi _t \; =\; \sum _{i \in \mathrm{Fin}\, M} w_t(i). \]
Lemma 111.20 Prefix Hedge weight is positive
#

For any game \(G\), learning rate \(\eta \), prefix of column actions, and row \(i\), the prefix Hedge weight \(w_t(i) = \exp (-\eta L_t(i)) \gt 0\).

Lemma 111.21 Prefix potential is positive

When \(M \ge 1\), the prefix potential \(\Phi _t \gt 0\), since it is a non-empty sum of positive terms.

Definition 111.22 Prefix Hedge mixed strategy

The normalised Hedge distribution after a prefix of \(t\) column actions is the mixed row strategy with weights

\[ p_t(i) \; =\; \frac{w_t(i)}{\Phi _t}, \]

where \(w_t(i)\) is the prefix Hedge weight and \(\Phi _t\) the prefix potential.

Definition 111.23 Online column best responses to Hedge

The sequence of column best responses generated online against Hedge: at time \(t\), given the history \(a_0, \dots , a_{t-1}\) of prior column choices, the column player responds with \(a_t = \mathrm{bestColumn}(G, p_t)\), where \(p_t\) is the prefix Hedge mixed strategy. This is defined by well-founded recursion on \(t\).

Lemma 111.24 Prefix loss equals cumulative loss

For any fixed column-action sequence \(a : \mathrm{Fin}\, T \to \mathrm{Fin}\, N\), round \(t \lt T\), and row \(i\), the prefix cumulative loss of row \(i\) along the first \(t\) actions equals the abstract cumulative loss of \(i\) in the induced loss sequence: \(\mathrm{prefixGameLoss}(\dots ) = \mathrm{cumLoss}(G.\mathrm{toLossSeq}(a), t, i)\).

Lemma 111.25 Prefix weight equals Hedge weight

The prefix Hedge weight of row \(i\) at round \(t\) equals the abstract Hedge weight \(\mathrm{hedgeWeight}(\eta , G.\mathrm{toLossSeq}(a), t, i)\) for the induced loss sequence.

Lemma 111.26 Prefix potential equals Hedge potential

The prefix potential at round \(t\) equals the abstract Hedge potential \(\mathrm{potential}(\eta , G.\mathrm{toLossSeq}(a), t)\) for the induced loss sequence.

Lemma 111.27 Prefix strategy weight equals Hedge distribution

The weight of row \(i\) in the prefix Hedge mixed strategy at round \(t\) equals the abstract Hedge distribution \(\mathrm{hedgeDist}(\eta , G.\mathrm{toLossSeq}(a), t, i)\).

Lemma 111.28 Regret-to-payoff bridge
#

Suppose the row player uses per-round distributions \(p_0, \dots , p_{T-1}\) (each summing to \(1\)) against column responses \(j_0, \dots , j_{T-1}\), and achieves cumulative loss regret at most \(R\):

\[ \sum _t \sum _i p_t(i)(1 - A_{i,j_t}) - \min _i \sum _t (1 - A_{i,j_t}) \; \le \; R. \]

Then the total payoff satisfies

\[ \sum _t \sum _i p_t(i) A_{i,j_t} \; \ge \; \sup _i \sum _t A_{i,j_t} - R. \]
Lemma 111.29 Average of per-round minima is at most minimum of averages
#

For any \(f : \mathrm{Fin}\, T \to \mathrm{Fin}\, N \to \mathbb {R}\) and \(T \gt 0\),

\[ \frac{1}{T} \sum _{t=0}^{T-1} \min _{j} f(t, j) \; \le \; \min _{j} \frac{1}{T} \sum _{t=0}^{T-1} f(t, j). \]

Given a game \(G\) with \(M \ge 2\) rows, a number of rounds \(T \gt 0\), and a learning rate \(\eta \gt 0\), there exist mixed strategies \(p\) (row) and \(q\) (column) such that for all pure \(i \in \mathrm{Fin}\, M\) and \(j \in \mathrm{Fin}\, N\),

\[ \mathrm{payoffVsPure}(G, p, j) + \frac{\ln M / \eta + \eta T / 8}{T} \; \ge \; \mathrm{pureVsPayoff}(G, i, q). \]

The strategies are constructed by running Hedge for \(T\) rounds, with the column player taking pure best responses, and then averaging.

Theorem 111.31 Approximate minimax

For any finite zero-sum game \(G\) with \(M \ge 2\) rows and any \(\varepsilon \gt 0\), there exist mixed strategies \(p\) (row) and \(q\) (column) forming an \(\varepsilon \)-approximate saddle point: for all pure \(i \in \mathrm{Fin}\, M\) and \(j \in \mathrm{Fin}\, N\),

\[ \mathrm{payoffVsPure}(G, p, j) + \varepsilon \; \ge \; \mathrm{pureVsPayoff}(G, i, q). \]
Theorem 111.32 Approximate minimax theorem

For any finite zero-sum game \(G\) with at least \(2\) row actions and any \(\varepsilon \gt 0\), there exist mixed strategies \(p, q\) such that for all pure \(i, j\),

\[ \mathrm{payoffVsPure}(G, p, j) + \varepsilon \; \ge \; \mathrm{pureVsPayoff}(G, i, q). \]

This is the same guarantee as approx_minimax, stated in the \(\forall \varepsilon \gt 0, \exists p, q\) form.