#include <bits/stdc++.h>
#pragma GCC optimize(2)
using namespace std;
int head[1205], to[100004], w[100004], nxt[100004], n, m, s, t, ccnt, cur[1205], a[1205], ans;
int maxl;
bool vis[1205];
int d[1205];
queue<int> que;
inline void add(int p, int q, int ww) {
	nxt[++ccnt] = head[p];
	to[ccnt] = q;
	w[ccnt] = ww;
	head[p] = ccnt;
}
inline bool bfs() {
	for (int i = 1; i <= n; i++) {
		cur[i] = head[i];
		d[i] = 0x3f3f3f3f;
	}
	d[s] = 0;
	while (que.size()) que.pop();
	que.push(s);
	int op;
	while (que.size()) {
		op = que.front();
		que.pop();
		for (int i = head[op]; i; i = nxt[i]) {
			if (w[i]   && d[to[i]] > d[op] + 1) {
				que.push(to[i]);
				d[to[i]] = d[op] + 1;
			}
		}
	}
	return (d[t] != 0x3f3f3f3f);
}
inline int dfs(int p, int q) {
	if (p == t) {
		maxl += q;
		return q;
	}
	int op = 0, r;
	for (int &i = cur[p]; i; i = nxt[i]) {
		if (w[i] > 0 && d[to[i]] == d[p] + 1) {
			r = dfs(to[i], min(w[i], q));
			if (r) {
				q -= r;
				op += r;
				w[i] -= r;
				w[i ^ 1] += r;
			}
			if (q == 0) break;
		}
	}
	return op;
}
inline int Dinic() {
	maxl = 0;
	while (bfs()) {
		dfs(s, 0x3f3f3f3f);
	}
	return maxl;
}
inline void ad(int p, int q, int ww) {
	add(p, q, ww);
	add(q, p, 0);
}
signed main() {
	ios::sync_with_stdio(0);
	cin.tie(0);
	cout.tie(0);
	cin >> n >> m ;
	ccnt = 1;
	int p, q, ww;
	s = 2 * n + 1, t = 2 * n + 2;
	for (int i = 1; i <= n; i++) {
		cin >> a[i];
		if (a[i] > 0) ad(s, i, a[i]), ad(i + n, t, a[i]), ans += a[i];
		else ad(i, i + n, -a[i]);
	}
	for (int i = 1; i <= m; i++) {
		cin >> p >> q;
		ad(p, q, 0x3f3f3f3f);
		ad(p + n, q + n, 0x3f3f3f3f);
	}
	n = 2 * n + 2;
	cout << ans - Dinic();
}
//网络流建模有点恶心
//一个负数前,后要考虑还要不要

//不管怎么建模都会一次流干

//5 4
//-99 100 -99 100
//2 1
//4 3
//1 5
//3 5