Fairly Allocating Goods, Chores and Mixed - Ruta Mehta

  • Intro
    • Chores appear in our daily life as frequently as goods, so chore allocation is just as important as allocating goods
    • we want to allocate both chores(negative) and goods(positive) and have a fair division
    • Assets vs liability and defence budgets vs duties is another example
  • Model
    • A set of agents
    • A set of divisible items
  • Competitive (market) Equilibrium CE
    • Optimal bundle refers to whatever the agents want to buy disregarding the supply constraints
    • The collection of optimal bundle is demand
    • and CE is when demand = supply
    • CEEI (CE equal income) is when all agents have the same budget
      • This is envy free and everything is allocated
      • Also we can get pareto optimal and nash welfare for a wide range of functions
      • This only words when items are divisible.
    • Computation of CE is extremely important apparently
  • CE with Bads/Chores
    • Prices are now payments
    • Agents have earning requirements
    • both the things together you we have a bundle
    • bundles together create a demand and now we minimise pain instead of maximising uhh happy and CE is when we have demand = supply
    • Equal budget corresponds to equal income
  • 1-homogeneous
    • value of a scaled bundle is the same as scaled value of the bundle
    • Even with CEEI with additive vals, bad only does not give a non convex set.
    • With number of agents and number of items is a constant, we have poly time algorithm
  • Approximation Algorithm - exterior point method
    • Bads W linear
      • A: set of agents who need to earn atleast 1 and linear valuation
      • M; set of divisible bad chores
      • feasable means demand = supply
      • Nash Welfare: Minimize the product of disutilities given the constraint that every variable is positive
      • But this is not a convex set so we only go after a local optima as global optima would be hard to find
      • Local optima condition is normal to the tangent hyperplane is equal to the gradient
      • Naive idea: gradient descent, always goes to the boundry where it should not go. because that is where the minima is but we disqualify it by extra constraints
      • Idea: Pick a point in the exterior, go to the closest point.
      • Then we go the nearest valid point.
      • Then we move along the tangent hyperplane and find the best point in it. this will likely be another exterior point but it will be objectively beter.
      • If we converge we are sure that we are at a solution and we converge in polynomial time.
  • She looks like he knows about her field very well