1 Introduction
Exploration is a long standing problem in Reinforcement Learning (RL), where optimism in the face of uncertainty is one fundamental principle (Lai & Robbins, 1985; Strehl & Littman, 2005). Here the uncertainty refers to parametric uncertainty
, which arises from the variance in the estimates of certain parameters given finite samples. Both countbased methods
(Auer, 2002; Kaufmann et al., 2012; Bellemare et al., 2016; Ostrovski et al., 2017; Tang et al., 2017) and Bayesian methods (Kaufmann et al., 2012; Chen et al., 2017; O’Donoghue et al., 2017) follow this optimism principle. In this paper, we propose to use distributional RL methods to achieve this optimism.Different from classical RL methods, where an expectation of value function is learned (Sutton, 1988; Watkins & Dayan, 1992; Mnih et al., 2015), distributional RL methods (Jaquette, 1973; Bellemare et al., 2017) maintain a full distribution of future return. In the limit, distributional RL captures the intrinsic uncertainty of an MDP (Bellemare et al., 2017; Dabney et al., 2017, 2018; Rowland et al., 2018). Intrinsic uncertainty arises from the stochasticity of the environment
, which is parameter and sample independent. However, it is not trivial to quantify the effects of parametric and intrinsic uncertainties in distribution learning. To investigate this, let us look closer at a simple setup of distribution learning. Here we use Quantile Regression (QR) (detailed in Section 2.2), but the example presented here holds for other distribution learning methods. Here the random samples are drawn from any stationary distribution. The initial estimated distribution is set to be the uniform one (left plots). At each time step, QR updates its estimate in an online fashion by minimizing some loss function. In the limit the estimated QR distribution converges to the true distribution (right plots). The two middle plots examine the intermediate estimated distributions before convergence in two distinct cases.
Case 1: Figure 0(a) shows a deterministic environment where the data is generated by a degenerate distribution. In this case, the intermediate estimate of the distribution (middle plot) contains only the information about parametric uncertainty. Here, parametric uncertainty comes from the error in the estimation of the quantiles. The left subplot shows estimation from the initialized parameters for the distribution estimator. The middle subplot shows the estimated distribution converges closer to the true distribution on the right subplot.
Case 2: Figure 0(b) shows a stochastic environment, where the data is generated by a nondegenerate (stationary) distribution. In this case, the intermediate estimated distribution is the result of both parametric and intrinsic uncertainties. In the middle plot, the distribution estimator (QR) models randomness from both parametric and intrinsic uncertainties, and it is hard to split them. The parametric uncertainty does go away over time and converge to the true distribution on the right subplot. Our main insight in this paper is that the upper bound for a stateaction value estimate shrinks at a certain rate (See Section 3
for details). Specifically, the error of the quantile estimator is known to converge asymptotically in distribution to the Normal distribution
(Koenker, 2005). By treating the estimated distribution during learning as subnormal we can estimate the upper bound of the stateaction values with a high confidence (by applying Hoeffding’s inequality).This example illustrates distributions learned via distributional methods (such as distributional RL algorithms) model the randomness arising from both intrinsic and parametric uncertainties. In this paper, we study how to take advantage of distributions learned by distributional RL methods for efficient exploration in the face of uncertainty.
To be more specific, we use Quantile Regression DeepQNetwork (QRDQN, (Dabney et al., 2017)) to learn the distribution of value function. We start with an examination of the two uncertainties and a naive solution that leaves the intrinsic uncertainty unsupressed. We construct a counter example in which this naive solution fails to learn. The intrinsic uncertainty persists and leads the naive solution to favor actions with higher variances. To suppress the intrinsic uncertainty, we apply a decaying schedule to improve the naive solution.
One interesting finding in our experiments is that the distributions learned by QRDQN can be asymmetric. By using the upper quantiles of the estimated distribution (Mullooly, 1988), we estimate an optimistic exploration bonus for QRDQN.
We evaluated our algorithm in 49 Atari games (Bellemare et al., 2013). Our approach achieved 483 % average gain in cumulative rewards over QRDQN. The overall improvement is reported in Figure 10.
We also compared our algorithm with QRDQN in a challenging 3D driving simulator (CARLA). Results show that our algorithm achieves nearoptimal safety rewards twice faster than QRDQN.
In the rest of this paper, we first present some preliminaries of RL Section 2. In Section 3, we then study the challenges posed by the mixture of parametric and intrinsic uncertainties, and propose a solution to suppress the intrinsic uncertainty. We also propose a truncated variance estimation for exploration bonus in this section. In Section 4, we present empirical results in Atari games. Section 5 contains results on CARLA. Section 6 an overview of related work, and Section 7 contains conclusion.
2 Background
2.1 Reinforcement Learning
We consider a Markov Decision Process (MDP) of a state space
, an action space , a reward “function” , a transition kernel , and a discount ratio . In this paper we treat the reward “function”as a random variable to emphasize its stochasticity. Bandit setting is a special case of the general RL setting, where we usually only have one state.
We use to denote a stochastic policy. We use to denote the random variable of the sum of the discounted rewards in the future, following the policy and starting from the state and the action . We have , where and . The expectation of the random variable is
which is usually called the stateaction value function. In general RL setting, we are usually interested in finding an optimal policy , such that holds for any . All the possible optimal policies share the same optimal stateaction value function , which is the unique fixed point of the Bellman optimality operator (Bellman, 2013),
Based on the Bellman optimality operator, Watkins & Dayan (1992) proposed Qlearning to learn the optimal stateaction value function for control. At each time step, we update as
where is a step size and is a transition. There have been many work extending Qlearning to linear function approximation (Sutton & Barto, 2018; Szepesvári, 2010). Mnih et al. (2015)
combined Qlearning with deep neural network function approximators, resulting the DeepQNetwork (DQN). Assume the
function is parameterized by a network, at each time step, DQN performs a stochastic gradient descent to update
minimizing the losswhere is target network (Mnih et al., 2015), which is a copy of and is synchronized with periodically, and is a transition sampled from a experience replay buffer (Mnih et al., 2015), which is a firstinfirstout queue storing previously experienced transitions. Decorrelating representation has shown to speed up DQN significantly (Mavrin et al., 2019a). For simplicity, in this paper we will focus on the case without decorrelation.
2.2 Quantile Regression
The core idea behind QRDQN is the Quantile Regression introduced by the seminal paper (Koenker & Bassett Jr, 1978)
. This approach gained significant attention in the field of Theoretical and Applied Statistics and might not be well known in other fields. For that reason we give a brief introduction here. Let us first consider QR in the supervised learning. Given data
, we want to compute the quantile of corresponding the quantile level . linear quantile regression loss is defined as:(1) 
where
(2) 
is the weighted sum of residuals. Weights are proportional to the counts of the residual signs and order of the estimated quantile . For higher quantiles positive residuals get higher weight and vice versa. If , then the estimate of the median for is , with .
2.3 Distributional RL
Instead of learning the expected return , distributional RL focuses on learning the full distribution of the random variable directly (Jaquette, 1973; Bellemare et al., 2017; Mavrin et al., 2019b). There are various approaches to represent a distribution in RL setting (Bellemare et al., 2017; Dabney et al., 2018; BarthMaron et al., 2018). In this paper, we focus on the quantile representation (Dabney et al., 2017) used in QRDQN, where the distribution of is represented by a uniform mix of supporting quantiles:
where denote a Dirac at , and each is an estimation of the quantile corresponding to the quantile level (a.k.a. quantile index) with for . The stateaction value is then approximated by . Such approximation of a distribution is referred to as quantile approximation.
Similar to the Bellman optimality operator in meancentered RL, we have the distributional Bellman optimality operator for control in distributional RL,
Based on the distributional Bellman optimality operator, Dabney et al. (2017) proposed to train quantile estimations (i.e., ) via the Huber quantile regression loss (Huber, 1964). To be more specific, at time step the loss is
where and
, where is the indicator function and is the Huber loss,
3 Algorithm
In this section we present our method. First, we study the issue of the mixture of parametric and intrinsic uncertainties in the estimated distributions learned by QR approach. We show that the intrinsic uncertainty has to be suppressed in calculating exploration bonus and introduce a decaying schedule to achieve this.
Second, in a simple example where the distribution is asymmetric, we show exploration bonus from truncated variance outperforms bonus from the variance. In fact, we did find that the distributions learned by QRDQN (in Atari games) can be asymmetric. Thus we combine the truncated variance for exploration in our method.
3.1 The issue of intrinsic uncertainty
A naive approach to exploration would be to use the variance of the estimated distribution as a bonus. We provide an illustrative counter example. Consider a multiarmed bandit environment with 10 arms where each arm’s reward follows normal distribution . In each run, means
are drawn from standard normal. Standard deviation of the best arm is set to 1.0, other arms’ standard deviations are set to 5. In the setting of multiarmed bandits, this approach leads to picking the arm
such that(3) 
where and are the estimated mean and variance of the th arm, computed from the corresponding quantile distribution estimation.
Figure 3 shows that naive exploration bonus fails. Figure 1(a) illustrates the reason for the failure of naive exploration bonus. The estimated QR distribution is a mixture of parametric and intrinsic uncertainties. Recall, as learning progresses the parametric uncertainty vanishes and the intrinsic uncertainty stays (Figure 1(b)). Therefore, this naive exploration bonus will tend to be biased towards intrinsic variation, which hurts performance. Note that the best arm has a low intrinsic variation. It is not chosen since its exploration bonus term is much smaller than the other arms as parametric uncertainty vanishes in all arms.
The major obstacle in using the estimated distribution by QR for exploration is the composition of parametric and intrinsic uncertainties, whose variance is measured by the term in (3). To suppress the intrinsic uncertainty, we propose a decaying schedule in the form of a multiplier to :
(4) 
Figure 1(b) depicts the exploration bonus resulting from the application of decaying schedule. From the classical QR theory (Koenker, 2005), it is known that the parametric uncertainty decays at the following rate:
(5) 
where is a constant factor.
We apply this new schedule to the counter example where the naive solution fails. As shown in Figure 3, this decaying schedule significantly outperforms the naive exploration bonus.
3.2 Assymetry and truncated variance
QR has no restriction on the family of distributions it can represent. In fact, the learned distribution can be asymmetric, defined by mean median. From Figure 5 it can be seen that the distribution estimated by QRDQN1 is mostly asymmetric. At the end of training, agent achieved nearly maximum score. Hence, the distributions correspond to the nearoptimal policy, but they are not symmetric.
For the sake of the argument, consider a simple decomposition of the variance of the QR’s estimated distribution into the two terms: the Right Truncated and Left Truncated variances ^{1}^{1}1Note: Right truncation means dropping left part of the distribution with respect to the mean:
where is the Right Truncated Variance and is the right. To simplify notation we assume is an even number here. The Right Truncated Variance tells about lower tail variability and the Left Truncated Variance tells about upper tail variability. In general, the two variances are not equal. ^{2}^{2}2Consider discrete empirical distribution with support
and probability atoms
. If the distribution is symmetric, then the two are the same.The Truncated Variance is equivalent to the Tail Conditional Variance (TCV):
(6) 
defined in (Valdez, 2005). For instantiating optimism in the face of uncertainty, the upper tail variability is more relevant than the lower tail, especially if the estimated distribution is asymmetric (Valdez, 2005). Intuitively speaking, is more optimistic. is biased towards positive rewards. To increase stability, we use the left truncated measure of the variability, , based on the median rather than the mean due to its wellknown statistical robustness (Huber, 2011; Hampel et al., 2011):
(7) 
where ’s are th quantiles. By combining decaying schedule from (5) with from (7) we obtain a new exploration bonus for picking an action, which we call Decaying Left Truncated Variance (DLTV).
In order to empirically validate our new approach we employ a multiarmed bandits environment with asymmetrically distributed rewards. In each run the means of arms are drawn from standard normal distribution. The best arm’s reward follow . Other arms rewards follow . We compare the performance of both exploration methods in another, symmetric environment with rewards following the normal distribution centered at corresponding means (same as the asymmetric environment) with unit variance.
The results are presented in Figure 4. With asymmetric reward distributions, the truncated variance exploration bonus significantly outperforms the naive variance exploration bonus. In addition, the performance of truncated variance is slightly better in the symmetric case.
3.3 DLTV for Deep RL
So far, we introduced the decaying schedule to control the parametric part of the composite uncertainty. Additionally, we introduced a truncated variance to improve performance in environments with asymmetric distributions. These ideas generalize in a straightforward fashion to the Deep RL setting. Algorithm 1 outlines DLTV for Deep RL. Action selection step in line 2 of Algorithm 1 uses exploration bonus in the form of defined in (7) and schedule defined in (5).
Figure 8 presents naive and decaying exploration bonus term from DLTV of QRDQN during training in Atari Pong. Comparison of Figure 8 to Figure 1(b) reveals the similarity in the behavior of the naive exploration bonus and the decaying exploration bonus. This shows what the raw variance looks like in Atari 2600 game and the suppressed intrinsic uncertainty leading to a decaying bonus as illustrated in Figure 1(b).
4 Atari 2600 Experiments
We evaluated DLTV on the set of 49 Atari games initially proposed by (Mnih et al., 2015). Algorithms were evaluated on 40 million frames^{3}^{3}3Equivalently, 10 million agent steps. 3 runs per game. The summary of the results is presented in Figure 10. Our approach achieved 483 % average gain in cumulative rewards ^{4}^{4}4The cumulative reward is a suitable performance measure for our experiments, since none of the learning curves exhibit plummeting behaviour. Plummeting is characterized by abrupt degradation of performance. In such cases the learning curve drops to the minimum and stays their indefinitely. A more detailed discussion of this point is presented in (Machado et al., 2017). over QRDQN1. Notably the performance gain is obtained in hard games such as Venture, PrivateEye, Montezuma Revenge and Seaquest. The median of human normalized performance reported in Figure 6 shows a significant improvement of DLTV over QRDQN1. We present learning curves for all 49 games in the Appendix.
The architecture of the network follows (Dabney et al., 2017). For our experiments we chose the Huber loss with ^{5}^{5}5QRDQN with is denoted as QRDQN1 in the work by (Dabney et al., 2017) due to its smoothness compared to loss of QRDQN0. (Smoothness is better suited for gradient descent methods). We followed closely (Dabney et al., 2017) in setting the hyper parameters, except for the learning rate of the Adam optimizer which we set to .
The most significant distinction of our DLTV is the way the exploration is performed. As opposed to QRDQN there is no epsilon greedy exploration schedule in DLTV. The exploration is performed via the term only (line 2 of Algorithm 1).
An important hyper parameter which is introduced by DLTV is the schedule, i.e. the sequence of multipliers for , . In our experiments we used the following schedule .
We studied the effect of the schedule in the Atari 2600 game Venture. Figure 7 show that constant schedule for DLTV significantly degenerates the performance. These empirical results show that the decaying schedule in DLTV is very important.
5 CARLA Experiments
A particularly interesting application of the (Distributional) RL approach is driving safety. There has been quite a converge of interests in using RL for autonomous driving, e.g., see (Sakib et al., 2019; Fridman et al., 2018; Chen et al., 2018; Yao et al., 2017). In the classical RL setting the agent only cares about the mean. In Distributional RL the estimate of the whole distribution allows for the construction of the risksensitive policies. For that reason we further validate DLTV in CARLA environment which is a 3D self driving simulator.
5.1 Sample efficiency
It should be noted that CARLA is a more visually complex environment than Atari 2600, since it is based on a modern Unreal Engine 4 with realistic physics and visual effects. For the purpose of this study we picked the task in which the ego car has to reach a goal position following predefined paths. In each episode the start and goal positions are sampled uniformly from a predefined set of locations (around 20). We conducted our experiments in Town 2. We simplified the reward signal provided in the original paper (Dosovitskiy et al., 2017). We assign reward of for any type of infraction and a a small positive reward for travelling in the correct direction without any infractions, i.e. . The infractions we consider are: collisions with cars, collisions with humans, collisions with static objects, driving on the opposite lane and driving on a sidewalk. The continuous action space was discretized in a coarse grain fashion. We defined 7 actions: 6 actions for going in different directions using fixed values for steering angle and throttle and a no op action. The training learning curves are presented in Figure 9. DLTV significantly outperforms QRDQN1 and DQN. Interestingly QRDQN1 performs on par with DQN.
5.2 Driving Safety
A byproduct of Distributional RL is the estimated distribution of . The access to this density allows for different approaches to control. For example Morimura et al. (2012) derive risksensitive policies based on the quantiles rather than the mean. The reasoning behind such approach is to view quantile as a risk metric. For instance, one particularly interesting risk metric is ValueatRisk (VaR) which has been in use for a few decades in Financial Industry (Philippe, 2006). Artzner et al. (1999) define as , that is quantile of .
It might be easier to understand the idea behind VaR in financial setting. Consider two investments: first investment will lose 1 dollar of its value or more with 10% probability () and second investment will lose 2 dollars or more of its value with 5 percent probability (). Second investment is riskier than the first one, that is a risksensitive investor will pick an investment with the higher VaR. This same reasoning applies directly to RL setting. Here, instead of investments we deal with actions. risksensitive policy will pick the action that has highest VaR. For instance Morimura et al. (2012) showed in a simple environment of Cliff Walk the policy maximizing low quantiles yields paths further away from the dangerous cliff.
Risksensitive policies are not only applicable to toy domains. In fact risk sensitive policies is a very important research question in selfdriving. In that respect CARLA is a non trivial domain where risksensitive policies can be thoroughly tested. In (Dosovitskiy et al., 2017) authors introduce simple safety performance metric such as average distance travelled between infractions. In addition to this metric we also consider the collision impact. This metric allows one to differentiate policies with the same average distance between infractions. Given the impact is not avoidable, a good policy should minimize the impact.
Average distance  or  Mean 
between infractions  
Opposite lane  4.55  1.35 
Sidewalk  None  None 
Collisionstatic  None  3.54 
Collisioncar  0.70  1.53 
Collisionpedestrian  52.33  16.41 
Average collision impact  
Collisionstatic  None  509.81 
Collisioncar  497.22  1078.76 
Collisionpedestrian  40.79  40.70 
Distance, km  104.69  98.66 
# of evaluation episodes  1000  1000 
We trained our agent using DLTV approach and during evaluation we used risksensitive policy derived from instead of the usual mean. Interestingly, this approach does employ meancentered RL at all. We benchmark this approach against the agent that uses mean for control. The safety results for the risksensitive and the mean agents are presented in Table 1. It can be seen that risksensitive agent significantly improves safety performance across almost all metrics, except for collisions with cars. However, the impact of colliding with cars is twice lower for the risksensitive agent.
6 Related Work
Tang & Agrawal (2018) combined Bayesian parameter updates with distributional RL for efficient exploration. However, they demonstrated improvement in only simple domains. Zhang et al. (2019) generated riskseeking and riskaverse policies via distributional RL for exploration, making use of both optimism and pessimism of intrinsic uncertainty. To our best knowledge, we are the first to use the parametric uncertainty in the estimated distributions learned by distributional RL algorithms for exploration.
For optimism in the face of uncertainty in deep RL setting, Bellemare et al. (2016) and Ostrovski et al. (2017) exploited a generative model to enable pseudocount. Tang et al. (2017) combined taskspecific features from an autoencoder with similarity hashing to count high dimensional states. Chen et al. (2017) used ensemble to compute variancebased exploration bonus. O’Donoghue et al. (2017) used uncertainty Bellman equation to propagate the uncertainty through time steps. Most of those approaches bring in nonnegligible computation overhead. In contrast, our DLTV achieves this optimism via distributional RL (QRDQN in particular) and requires very little extra computation.
7 Conclusions
Recent advancements in distributional RL, not only established new theoretically sound principles but also achieved stateoftheart performance in challenging high dimensional environments like Atari 2600. We take a step further by studying the learned distributions by QRDQN, and discovered the composite effect of intrinsic and parametric uncertainties is challenging for efficient exploration. In addition, the distribution estimated by distributional RL can be asymmetric. We proposed a novel decaying scheduling to suppress the intrinsic uncertainty, and a truncated variance for calculating exploration bonus, resulting in a new exploration strategy for QRDQN. Empirical results showed that the our method outperforms QRDQN (with epsilongreedy strategy) significantly in Atari 2600. Our method can be combined with other advancements in deep RL, e.g. Rainbow (Hessel et al., 2017), to yield yet better results.
References
 Artzner et al. (1999) Artzner, P., Delbaen, F., Eber, J.M., and Heath, D. Coherent measures of risk. Mathematical finance, 9(3):203–228, 1999.

Auer (2002)
Auer, P.
Using confidence bounds for exploitationexploration tradeoffs.
Journal of Machine Learning Research
, 3(3):397–422, 2002.  BarthMaron et al. (2018) BarthMaron, G., Hoffman, M. W., Budden, D., Dabney, W., Horgan, D., Muldal, A., Heess, N., and Lillicrap, T. Distributed distributional deterministic policy gradients. arXiv:1804.08617, 2018.
 Bellemare et al. (2016) Bellemare, M., Srinivasan, S., Ostrovski, G., Schaul, T., Saxton, D., and Munos, R. Unifying countbased exploration and intrinsic motivation. NIPS, 2016.

Bellemare et al. (2013)
Bellemare, M. G., Naddaf, Y., Veness, J., and Bowling, M.
The arcade learning environment: An evaluation platform for general
agents.
Journal of Artificial Intelligence Research
, 47:253–279, 2013.  Bellemare et al. (2017) Bellemare, M. G., Dabney, W., and Munos, R. A distributional perspective on reinforcement learning. arXiv:1707.06887, 2017.
 Bellman (2013) Bellman, R. Dynamic programming. Courier Corporation, 2013.
 Chen et al. (2018) Chen, C., Qian, J., Yao, H., Luo, J., Zhang, H., and Liu, W. Towards comprehensive maneuver decisions for lane change using reinforcement learning. NIPS Workshop on Machine Learning for Intelligent Transportation Systems (MLITS), 2018.
 Chen et al. (2017) Chen, R. Y., Sidor, S., Abbeel, P., and Schulman, J. Ucb exploration via qensembles. arXiv:1706.01502, 2017.
 Dabney et al. (2017) Dabney, W., Rowland, M., Bellemare, M. G., and Munos, R. Distributional reinforcement learning with quantile regression. arXiv:1710.10044, 2017.
 Dabney et al. (2018) Dabney, W., Ostrovski, G., Silver, D., and Munos, R. Implicit quantile networks for distributional reinforcement learning. arXiv:1806.06923, 2018.
 Dosovitskiy et al. (2017) Dosovitskiy, A., Ros, G., Codevilla, F., Lopez, A., and Koltun, V. Carla: An open urban driving simulator. arXiv:1711.03938, 2017.
 Fridman et al. (2018) Fridman, L., Jenik, B., and Terwilliger, J. Deeptraffic: Driving fast through dense traffic with deep reinforcement learning. arxiv:1801.02805, 2018.
 Hampel et al. (2011) Hampel, F. R., Ronchetti, E. M., Rousseeuw, P. J., and Stahel, W. A. Robust statistics: the approach based on influence functions, volume 196. John Wiley & Sons, 2011.
 Hessel et al. (2017) Hessel, M., Modayil, J., Van Hasselt, H., Schaul, T., Ostrovski, G., Dabney, W., Horgan, D., Piot, B., Azar, M., and Silver, D. Rainbow: Combining improvements in deep reinforcement learning. arXiv:1710.02298, 2017.
 Huber (1964) Huber, P. J. Robust estimation of a location parameter. The Annals of Mathematical Statistics, 35(1):73–101, 1964.
 Huber (2011) Huber, P. J. Robust statistics. International Encyclopedia of Statistical Science, 35(1):1248–1251, 2011.
 Jaquette (1973) Jaquette, S. C. Markov decision processes with a new optimality criterion: Discrete time. The Annals of Statistics, 1(3):496–505, 1973.
 Kaufmann et al. (2012) Kaufmann, E., Cappé, O., and Garivier, A. On bayesian upper confidence bounds for bandit problems. AISTAT, 2012.
 Koenker (2005) Koenker, R. Quantile Regression. Econometric Society Monographs. Cambridge University Press, 2005.
 Koenker & Bassett Jr (1978) Koenker, R. and Bassett Jr, G. Regression quantiles. Econometrica: Journal of the Econometric Society, 46(1):33–50, 1978.
 Lai & Robbins (1985) Lai, T. L. and Robbins, H. Asymptotically efficient adaptive allocation rules. Advances in Applied Mathematics, 6(1):4–22, 1985.
 Machado et al. (2017) Machado, M. C., Bellemare, M. G., Talvitie, E., Veness, J., Hausknecht, M., and Bowling, M. Revisiting the arcade learning environment: Evaluation protocols and open problems for general agents. arXiv:1709.06009, 2017.
 Mavrin et al. (2019a) Mavrin, B., Yao, H., and Kong, L. Deep reinforcement learning with decorrelation. arxiv:1903.07765, 2019a.
 Mavrin et al. (2019b) Mavrin, B., Zhang, S., Yao, H., and Kong, L. Exploration in the face of parametric and intrinsic uncertainties. AAMAS, 2019b.
 Mnih et al. (2015) Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al. Humanlevel control through deep reinforcement learning. Nature, 518(7540):529, 2015.
 Morimura et al. (2012) Morimura, T., Sugiyama, M., Kashima, H., Hachiya, H., and Tanaka, T. Parametric return density estimation for reinforcement learning. arXiv:1203.3497, 2012.
 Mullooly (1988) Mullooly, J. P. The variance of lefttruncated continuous nonnegative distributions. The American Statistician, 42(3):208–210, 1988.
 O’Donoghue et al. (2017) O’Donoghue, B., Osband, I., Munos, R., and Mnih, V. The uncertainty bellman equation and exploration. arXiv:1709.05380, 2017.
 Ostrovski et al. (2017) Ostrovski, G., Bellemare, M. G., Oord, A. v. d., and Munos, R. Countbased exploration with neural density models. arXiv:1703.01310, 2017.
 Philippe (2006) Philippe, J. Value at risk: the new benchmark for managing financial risk, 3rd Ed. McGrawHill Education, 2006.
 Rowland et al. (2018) Rowland, M., Bellemare, M. G., Dabney, W., Munos, R., and Teh, Y. W. An analysis of categorical distributional reinforcement learning. arXiv:1802.08163, 2018.
 Sakib et al. (2019) Sakib, N., Yao, H., and Zhang, H. Reinforcing classical planning for adversary driving scenarios. arxiv:1903.08606, 2019.
 Strehl & Littman (2005) Strehl, A. L. and Littman, M. L. A theoretical analysis of modelbased interval estimation. ICML, 2005.
 Sutton (1988) Sutton, R. S. Learning to predict by the methods of temporal differences. Machine Learning, 3(1):9–44, 1988.
 Sutton & Barto (2018) Sutton, R. S. and Barto, A. G. Reinforcement learning: An introduction (2nd Edition). MIT press, 2018.
 Szepesvári (2010) Szepesvári, C. Algorithms for Reinforcement Learning. Morgan and Claypool, 2010.
 Tang et al. (2017) Tang, H., Houthooft, R., Foote, D., Stooke, A., Chen, O. X., Duan, Y., Schulman, J., DeTurck, F., and Abbeel, P. # exploration: A study of countbased exploration for deep reinforcement learning. NIPS, 2017.
 Tang & Agrawal (2018) Tang, Y. and Agrawal, S. Exploration by distributional reinforcement learning. arXiv:1805.01907, 2018.
 Valdez (2005) Valdez, E. A. Tail conditional variance for elliptically contoured distributions. Belgian Actuarial Bulletin, 5(1):26–36, 2005.
 Watkins & Dayan (1992) Watkins, C. J. and Dayan, P. Qlearning. Machine Learning, 8(34):279–292, 1992.
 Yao et al. (2017) Yao, H., Nosrati, M. S., and Rezaee, K. Montecarlo tree search vs. modelpredictive controller: A trackfollowing example. NIPS Workshop on Machine Learning for Intelligent Transportation Systems (MLITS), 2017.
 Zhang et al. (2019) Zhang, S., Mavrin, B., Yao, H., Kong, L., and Liu, B. QUOTA: The quantile option architecture for reinforcement learning. AAAI, 2019.
Acknowledgement
The correct author list for this paper is Borislav Mavrin, Shangtong Zhang, Hengshuai Yao, Linglong Kong, Kaiwen Wu and Yaoliang Yu. Due to time pressure, Shangtong’s name was forgotten during submiting. If you cite this paper, please use this correct author list. The mistake was fixed in the arxiv version of this paper.
Appendix A Performance Profiling on Atari Games
Figure 10 shows the performance of DLTV and QRDQN on 49 Atari games, which is measured by cumulative rewards (normalized Area Under the Curve).