arXiv subjects

Pahan Dewasurendra

Publications and source records attributed to Pahan Dewasurendra.

8 recordsLinked to original sources

arXiv2026

Two Dimensions Govern Agnostic Multiclass Transductive Learning

In transductive classification, an adversary fixes a labeled population, one label is hidden uniformly, and the learner sees all remaining labels. For binary classes, agnostic transductive and PAC learning have the same minimax rate. Whether this extends to multiclass learning was open, especially for unbounded label spaces where uniform convergence can fail. We resolve the question up to logarithmic factors. For every multiclass class $\mathcal H$ with DS dimension $d_{DS}$ and Natarajan dimension $d_{\mathrm N}$, the optimal agnostic transductive excess error satisfies $\widetilde\Theta\left(\frac{d_{DS}}{n}+\sqrt{\frac{d_{\mathrm N}}{n}}\right).$ The result holds for arbitrary label spaces. The two terms are both necessary. A DS pseudo-cube gives the realizable $d_{DS}/n$ obstruction, while a Natarajan cube with repeated points and fair labels gives the agnostic $\sqrt{d_{\mathrm N}/n}$ obstruction. The upper bound uses a random-reservation principle. The learner deliberately ignores a constant fraction of the visible labels, which makes the true test point uniform in a large unseen block. We combine realizable compression, a label-space reduction, and inside-menu agnostic compression across this finite-population split. A new without-replacement multiplicative-weights lemma preserves the fast $d_{DS}/n$ term. Consequently, agnostic multiclass PAC and transductive learning obey the same two-dimension law up to logarithmic factors.

cs.LG↗

arXiv2026

The Sharp Tail of Uniform Stability

Uniform stability controls how much one training example can change the loss at any test point. A new logarithmic-free upper bound shows that a $\gamma$-uniformly stable algorithm with loss in $[0,L]$ has generalization gap at most $O \left(\gamma\log(1/\delta) +L\sqrt{\frac{\log(1/\delta)}{n}}\right)$ with probability $1-\delta$. Whether an actual bounded-loss learning algorithm can realize the linear dependence on $\log(1/\delta)$ has remained open. The known construction realizes it only for auxiliary weakly dependent random variables whose pointwise range grows with $n$. The known learning lower bound holds only at constant probability. We close this gap. For every $n$, stability level $\gamma$, and loss bound $L$, we construct one deterministic $\gamma$-uniformly stable learning problem whose tail satisfies, simultaneously for $1\le p\le c n$, $\mathbb P \left( R(A_S)-R_S(A_S) \ge c'\min \left\{L,\gamma p+L\sqrt{p/n}\right\} \right)\ge e^{-p}.$ The construction is ordinary bounded absolute-loss regression with constant labels. Its key is a multiscale collection of rare Rademacher features. A coordinatewise ramp is stable in sup norm, while an odd symmetrized maximum converts a unique extreme feature into a gap of order $\gamma p$ without violating the loss bound. Geometrically spaced ramps put all confidence levels into the same problem. Together with the logarithmic-free upper bound, this determines the optimal high-probability and moment dependence of uniform stability up to universal constants.

cs.LG↗

arXiv2026

Self-Bounding Regret Matching+ in Potential Games and Product-Simplex Optimization

Regret matching+ (RM+) is parameter free, scale invariant, and central to large game solving, but its only general individual-regret guarantee grows as $\sqrt{T}$. A recent ICLR result used this envelope to prove that RM+ reaches an $\epsilon$-stationary point of a smooth objective over a product of simplices in $O(\epsilon^{-4})$ iterations, or $O(\epsilon^{-8})$ from the standard zero initialization. We give an exact one-step conservation law for RM+. It states that forward utility gain pays for both squared state motion and growth of the regret-state norm. Norm growth is at most $\sqrt{m-1}$ times forward gain for $m$ actions, and the coefficient is sharp. This yields four results for unmodified RM+. Its regret on any utility path is controlled by centered temporal variation. Its regret is uniformly bounded under alternating play in every finite exact potential game, resolving an open question and making squared activation gaps summable. Both certified lazy and ordinary cyclic play attain an $\epsilon^{-2}$ exponent. On any smooth, possibly nonconcave simplex objective, RM+ finds an $\epsilon$-KKT point in $O(\epsilon^{-2})$ iterations. Most broadly, for a smooth objective over an arbitrary product of simplices, cyclic block RM+ attains the same $O(\epsilon^{-2})$ exponent from arbitrary initialization, with an explicit trajectory-dependent constant. The proof controls the finite objective loss caused by low-state blocks and then self-bounds every block state and the total squared path length. Complete proofs cover zero states, sharpness, common-profile stationarity, and robust gain dominance. Oracle-normalized diagnostics compare RM+ with predictive and smooth extra-gradient variants on graphical potential games and dense nonconvex objectives.

cs.GT↗

arXiv2026

Convex Networks Remain Hard to Certify: Dimension-Accuracy Barriers for Lipschitz Constants

Input-convex neural networks permit globally tractable minimization over their inputs, so one might expect their global regularity to be tractable in low input dimension. We prove exact and accuracy-sensitive barriers to this expectation. Given a bias-free one-hidden-layer ReLU network $f(x)=\sum_{r=1}^n \mathrm{ReLU}(a_r^\top x)$ with unit positive output weights, deciding whether its global Euclidean Lipschitz constant is at least a rational threshold is NP-complete and W[1]-hard when parameterized by the input dimension $d$. The same holds on the unit ball and with integral first-layer weights having at most nine nonzeros. More sharply, no deterministic multiplicative approximation scheme runs in $g(d)\mathrm{poly}(\mathcal B,1/\varepsilon)$ time unless FPT equals W[1]. Under the Exponential Time Hypothesis, no such algorithm runs in $g(d)(\mathcal B+1/\varepsilon)^{o(d/\log d)}$ time. Thus accuracy cannot have a polynomial dependence separated from dimension. The exact result resolves the Euclidean case of an open problem posed at COLT 2025 and left open by the ICLR 2026 parameterized hardness theory for general two-layer networks. The approximation barrier is specific to generator-presented zonotopes, complementing known $(1/\varepsilon)^{O(d)}$-time schemes and an analogous barrier for halfspace-presented polytopes. Our lifted-selector reduction has an inverse-polynomial radial gap, proved through a quantitative theorem for rational cyclic zonogons. Equivalently, the results apply to Euclidean zonotope radius and positive-semidefinite binary quadratic maximization parameterized by rank. Convexity makes minimization easy, but it does not make global sensitivity fixed-parameter tractable or permit a dimension-separated fully polynomial accuracy guarantee.

cs.CC↗

arXiv2026

Dirichlet Follow-the-Leader Closes the Gap in Simultaneous Multiclass U-Calibration

Can one forecaster attain the optimal regret rate for every bounded proper loss and also adapt to every smooth proper loss? Recent work answered this up to a dimension gap. Its self-concordant perturbation gives roughly $K^{5/4}\sqrt{T}$ worst-case regret and incurs an additional $\beta\sqrt{K}\log K$ for $\beta$-smooth losses. We close both gaps with a one-line forecaster. After observing class counts $c_{t-1}$, draw the next prediction from $\operatorname{Dir}(c_{t-1})$, on the face of classes seen so far. This is a fresh Bayesian bootstrap of the outcomes. The analysis rests on an exact identity: averaging any bounded proper loss under $\operatorname{Dir}(\alpha)$ equals a discrete derivative of its Dirichlet-averaged Bayes risk. The identity makes the be-the-perturbed-leader term telescope to a nonpositive Jensen gap. A one-count likelihood ratio then bounds stability by the inverse square root of that class's count. The resulting single, horizon-free algorithm satisfies $\sup_{\ell}\mathbb{E}\operatorname{Reg}_{\ell}\leq 4\sqrt{S_T T}\leq 4\sqrt{K T}$ and $\mathbb{E}\operatorname{Reg}_{\ell}\leq \frac{5}{2}\beta(1+\log T)$ for every $\beta$-smooth proper loss. Here $S_T$ is the number of observed classes. Known lower bounds show that both rates are optimal in their nontrivial regimes. The proof covers nondifferentiable losses and changes of the active simplex face.

cs.LG↗

arXiv2026

Multiscale Reward Hedging from Correct Demonstrations

Learning from correct demonstrations is harder than supervised learning when many answers are correct: after predicting, the learner sees one valid answer but not whether its own answer was valid, nor any reward. Existing reward-hedging guarantees consequently assume a finite reward class. We give the first horizon-free guarantee for continuous classes. The key is to hedge in one shared vote over tolerant optimality tests at every accuracy scale. A target reward has one surviving proxy per scale, and a prediction with gap above that scale doubles the proxy. This yields the simultaneous tail bound $|\{t:\ell_t>2^{-j}\}|\leq \log_2\mathcal N(\mathcal G,2^{-j-1})+j$, where $\mathcal G$ is the class of optimality-gap functions. Integrating the tails gives cumulative hidden gap bounded by a metric-entropy integral, independently of the number of rounds. Polynomial entropy $(A/\epsilon)^d$ gives $O(d\log A)$ total gap and a fast $O(d/m)$ statistical rate. For bounded linear contextual recommendation, the result is $O(d)$ regret for arbitrary compact menus. This is the first polynomial finite bound without structural restrictions on the menus, at the price of improper prediction. Although the general vote can be expensive, it is exactly polynomial-time for one-dimensional Lipschitz parameter curves. Fixed-radius rank-two recommendation takes $O(KT^2)$ time for menus of size $K$. We also prove an $\Omega(d)$ lower bound, low-rank and bounded ReLU-network corollaries, and a robust theorem that adds only the demonstrator's cumulative suboptimality. A reproducible adaptive stress test illustrates the predicted scale adaptation. After factorization, an exact MovieLens audit runs in 1.7 CPU seconds across ten users and improves mean latent gap over both a demonstrated-rating policy and a proper online baseline. The learner uses only action demonstrations and never observes a reward or a loss.

cs.LG↗

arXiv2024

Unsupervised Replay Strategies for Continual Learning with Limited Data

Artificial neural networks (ANNs) show limited performance with scarce or imbalanced training data and face challenges with continuous learning, such as forgetting previously learned data after new tasks training. In contrast, the human brain can learn continuously and from just a few examples. This research explores the impact of 'sleep', an unsupervised phase incorporating stochastic activation with local Hebbian learning rules, on ANNs trained incrementally with limited and imbalanced datasets, specifically MNIST and Fashion MNIST. We discovered that introducing a sleep phase significantly enhanced accuracy in models trained with limited data. When a few tasks were trained sequentially, sleep replay not only rescued previously learned information that had been catastrophically forgetting following new task training but often enhanced performance in prior tasks, especially those trained with limited data. This study highlights the multifaceted role of sleep replay in augmenting learning efficiency and facilitating continual learning in ANNs.

cs.LG↗

arXiv2024

Sleep-Like Unsupervised Replay Improves Performance when Data are Limited or Unbalanced

The performance of artificial neural networks (ANNs) degrades when training data are limited or imbalanced. In contrast, the human brain can learn quickly from just a few examples. Here, we investigated the role of sleep in improving the performance of ANNs trained with limited data on the MNIST and Fashion MNIST datasets. Sleep was implemented as an unsupervised phase with local Hebbian type learning rules. We found a significant boost in accuracy after the sleep phase for models trained with limited data in the range of 0.5-10% of total MNIST or Fashion MNIST datasets. When more than 10% of the total data was used, sleep alone had a slight negative impact on performance, but this was remedied by fine-tuning on the original data. This study sheds light on a potential synaptic weight dynamics strategy employed by the brain during sleep to enhance memory performance when training data are limited or imbalanced.

cs.NE↗