-
Boosting as Frank-Wolfe
Authors:
Ryotaro Mitsuboshi,
Kohei Hatano,
Eiji Takimoto
Abstract:
Some boosting algorithms, such as LPBoost, ERLPBoost, and C-ERLPBoost, aim to solve the soft margin optimization problem with the $\ell_1$-norm regularization. LPBoost rapidly converges to an $ε$-approximate solution in practice, but it is known to take $Ω(m)$ iterations in the worst case, where $m$ is the sample size. On the other hand, ERLPBoost and C-ERLPBoost are guaranteed to converge to an…
▽ More
Some boosting algorithms, such as LPBoost, ERLPBoost, and C-ERLPBoost, aim to solve the soft margin optimization problem with the $\ell_1$-norm regularization. LPBoost rapidly converges to an $ε$-approximate solution in practice, but it is known to take $Ω(m)$ iterations in the worst case, where $m$ is the sample size. On the other hand, ERLPBoost and C-ERLPBoost are guaranteed to converge to an $ε$-approximate solution in $O(\frac{1}{ε^2} \ln \frac{m}ν)$ iterations. However, the computation per iteration is very high compared to LPBoost.
To address this issue, we propose a generic boosting scheme that combines the Frank-Wolfe algorithm and any secondary algorithm and switches one to the other iteratively. We show that the scheme retains the same convergence guarantee as ERLPBoost and C-ERLPBoost. One can incorporate any secondary algorithm to improve in practice. This scheme comes from a unified view of boosting algorithms for soft margin optimization. More specifically, we show that LPBoost, ERLPBoost, and C-ERLPBoost are instances of the Frank-Wolfe algorithm. In experiments on real datasets, one of the instances of our scheme exploits the better updates of the secondary algorithm and performs comparably with LPBoost.
△ Less
Submitted 30 September, 2022; v1 submitted 22 September, 2022;
originally announced September 2022.
-
The cohomology of framed moduli spaces and the coordinate ring of torus fixed points of quotient singularities
Authors:
Kohei Hatano
Abstract:
If two conical symplectic resolutions $X\to X_0$ and $X^!\to X_0^!$ are symplectic dual, the cohomology ring $H^*(X)$ and the coordinate ring of $\mathbb{C}^*$-fixed points in $X_0^!$ are expected to be isomorphic as graded algebras. This statement is called Hikita conjecture and it is known that the conjecture holds for some cases. In this paper, we deal with the cohomology of framed moduli space…
▽ More
If two conical symplectic resolutions $X\to X_0$ and $X^!\to X_0^!$ are symplectic dual, the cohomology ring $H^*(X)$ and the coordinate ring of $\mathbb{C}^*$-fixed points in $X_0^!$ are expected to be isomorphic as graded algebras. This statement is called Hikita conjecture and it is known that the conjecture holds for some cases. In this paper, we deal with the cohomology of framed moduli spaces over the projective plane and the coordinate ring of $\mathbb{C}^*$- fixed points of $\mathbb{C}^{2n}/((\mathbb{Z}/r\mathbb{Z})\wr S_n) $ and show that these are isomorphic as graded vector spaces.
△ Less
Submitted 5 October, 2021; v1 submitted 29 September, 2021;
originally announced September 2021.
-
Pusz--Woronowicz's functional calculus revisited
Authors:
Kanae Hatano,
Yoshimichi Ueda
Abstract:
This note is a complement to Pusz--Woronowicz's works on functional calculus for two positive forms from the viewpoint of operator theory. Based on an elementary, self-contained and purely Hilbert space operator explanation of their functional calculus, we show that any operator connection type operations (including any operator perspectives) are captured by their functional calculus.
This note is a complement to Pusz--Woronowicz's works on functional calculus for two positive forms from the viewpoint of operator theory. Based on an elementary, self-contained and purely Hilbert space operator explanation of their functional calculus, we show that any operator connection type operations (including any operator perspectives) are captured by their functional calculus.
△ Less
Submitted 31 August, 2021; v1 submitted 23 December, 2020;
originally announced December 2020.
-
A generalised log-determinant regularizer for online semi-definite programming and its applications
Authors:
Yaxiong Liu,
Ken-ichiro Moridomi,
Kohei Hatano,
Eiji Takimoto
Abstract:
We consider a variant of online semi-definite programming problem (OSDP): The decision space consists of semi-definite matrices with bounded $Γ$-trace norm, which is a generalization of trace norm defined by a positive definite matrix $Γ.$ To solve this problem, we utilise the follow-the-regularized-leader algorithm with a $Γ$-dependent log-determinant regularizer. Then we apply our generalised se…
▽ More
We consider a variant of online semi-definite programming problem (OSDP): The decision space consists of semi-definite matrices with bounded $Γ$-trace norm, which is a generalization of trace norm defined by a positive definite matrix $Γ.$ To solve this problem, we utilise the follow-the-regularized-leader algorithm with a $Γ$-dependent log-determinant regularizer. Then we apply our generalised setting and our proposed algorithm to online matrix completion(OMC) and online similarity prediction with side information. In particular, we reduce the online matrix completion problem to the generalised OSDP problem, and the side information is represented as the $Γ$ matrix. Hence, due to our regret bound for the generalised OSDP, we obtain an optimal mistake bound for the OMC by removing the logarithmic factor.
△ Less
Submitted 11 December, 2020; v1 submitted 10 December, 2020;
originally announced December 2020.
