//#define NDEBUG
#pragma warning(disable : 4146)

#include <bits/stdc++.h>

namespace n91 {

using i32 = std::int32_t;
using i64 = std::int64_t;
using u32 = std::uint32_t;
using u64 = std::uint64_t;
using isize = std::ptrdiff_t;
using usize = std::size_t;
using f64 = double;

struct rep {
  struct itr {
    usize i;
    constexpr itr(const usize i) noexcept : i(i) {}
    void operator++() noexcept { ++i; }
    constexpr usize operator*() const noexcept { return i; }
    constexpr bool operator!=(const itr x) const noexcept { return i != x.i; }
  };
  const itr f, l;
  constexpr rep(const usize f, const usize l) noexcept
      : f(std::min(f, l)), l(l) {}
  constexpr auto begin() const noexcept { return f; }
  constexpr auto end() const noexcept { return l; }
};
struct revrep {
  struct itr {
    usize i;
    constexpr itr(const usize i) noexcept : i(i) {}
    void operator++() noexcept { --i; }
    constexpr usize operator*() const noexcept { return i; }
    constexpr bool operator!=(const itr x) const noexcept { return i != x.i; }
  };
  const itr f, l;
  constexpr revrep(const usize f, const usize l) noexcept
      : f(l - 1), l(std::min(f, l) - 1) {}
  constexpr auto begin() const noexcept { return f; }
  constexpr auto end() const noexcept { return l; }
};
template <class T> auto md_vec(const usize n, const T &value) {
  return std::vector<T>(n, value);
}
template <class... Args> auto md_vec(const usize n, Args... args) {
  return std::vector<decltype(md_vec(args...))>(n, md_vec(args...));
}
template <class T> constexpr T difference(const T &a, const T &b) noexcept {
  return a < b ? b - a : a - b;
}
template <class T> void chmin(T &a, const T &b) noexcept {
  if (b < a)
    a = b;
}
template <class T> void chmax(T &a, const T &b) noexcept {
  if (a < b)
    a = b;
}
template <class F> class rec_lambda {
  F f;

public:
  rec_lambda(F &&f_) : f(std::forward<F>(f_)) {}
  template <class... Args> auto operator()(Args &&... args) const {
    return f(*this, std::forward<Args>(args)...);
  }
};
template <class T> T scan() {
  T ret;
  std::cin >> ret;
  return ret;
}
constexpr char eoln = '\n';

i64 floor_div(const i64 n, const i64 d) {
  assert(d != 0);
  return n / d - static_cast<i64>((n ^ d) < 0 && n % d != 0);
}

i64 ceil_div(const i64 n, const i64 d) {
  assert(d != 0);
  return n / d + static_cast<i64>((n ^ d) >= 0 && n % d != 0);
}

#ifdef N91_LOCAL
#define OJ_LOCAL(a, b) b
#else
#define OJ_LOCAL(a, b) a
#endif

} // namespace n91

#include <atcoder/modint>
using mint = atcoder::modint998244353;

struct dsu {
  std::vector<int> p;
  dsu(int n) : p(n, -1) {}
  int find(int i) { return p[i] == -1 ? i : p[i] = find(p[i]); }
  void merge(int x, int y) {
    x = find(x);
    y = find(y);
    p[y] = x;
  }
};

namespace n91 {

struct arcset {
  mint add;
  std::set<usize> st;
};

void main_() {
  const usize n = scan<usize>();
  const usize m = scan<usize>();
  auto g = md_vec(n, 0, usize());
  std::vector<arcset> as(n);
  for (const usize i : rep(0, m)) {
    const usize u = scan<usize>() - 1;
    const usize v = scan<usize>() - 1;
    g[v].push_back(u);
    as[u].st.insert(v);
  }

  dsu uf(n);
  std::vector<mint> dp(n);
  for (const usize v : rep(0, n)) {
    dp[v] += 1;
    for (const usize u : g[v]) {
      const usize x = uf.find(u);
      if (x == v)
        continue;
      uf.merge(v, x);
      if (as[x].st.erase(v))
        dp[v] += as[x].add;
      if (as[v].st.size() < as[x].st.size())
        std::swap(as[v], as[x]);
      for (const usize t : as[x].st) {
        dp[t] += as[x].add;
        if (as[v].st.insert(t).second)
          dp[t] -= as[v].add;
      }
      as[x].st.clear();
    }
    as[v].add += dp[v];
  }

  std::cout << std::accumulate(dp.begin(), dp.end(), mint(0)).val() << eoln;
}

} // namespace n91

int main() {
  //*
  std::ios::sync_with_stdio(false);
  std::cin.tie(nullptr);
  //*/
  std::cout << std::fixed << std::setprecision(20);
  n91::main_();
  return 0;
}