WEIGHTS STAGNATION IN DYNAMIC LOCAL SEARCH FOR SAT

Jan Zizka et al. (Eds) : CCSEIT, AIAP, DMDB, MoWiN, CoSIT, CRIS, SIGL, ICBB, CNSA-2016
pp. 79–90, 2016. © CS & IT-CSCP 2016 DOI : 10.5121/csit.2016.60607
WEIGHTS STAGNATION IN DYNAMIC
LOCAL SEARCH FOR SAT
Abdelraouf Ishtaiwi
Faculty of Information Technology, University of Petra, Amman, Jordan
faishtaiwig@uop.edu.jo
ABSTRACT
Since 1991, tries were made to enhance the stochastic local search techniques (SLS). Some
researchers turned their focus on studying the structure of the propositional satisfiability
problems (SAT) to better understand their complexity in order to come up with better
algorithms. Other researchers focused in investigating new ways to develop heuristics that alter
the search space based on some information gathered prior to or during the search process.
Thus, many heuristics, enhancements and developments were introduced to improve SLS
techniques performance during the last three decades. As a result a group of heuristics were
introduced namely Dynamic Local Search (DLS) that could outperform the systematic search
techniques. Interestingly, a common characteristic of DLS heuristics is that they all depend on
the use of weights during searching for satisfiable formulas.
In our study we experimentally investigated the weights behaviors and movements during
searching for satisfiability using DLS techniques, for simplicity, DDFW DLS heuristic is chosen.
As a results of our studies we discovered that while solving hard SAT problems such as blocks
world and graph coloring problems, weights stagnation occur in many areas within the search
space. We conclude that weights stagnation occurrence is highly related to the level of the
problem density, complexity and connectivity.
1. INTRODUCTION
The propositional satisfiability (SAT) problem is at the core of many computer science and
artificial intelligence problems. Hence, finding efficient solutions for SAT has far reaching
implications. In this study, we consider propositional formulae in conjunctive normal form
in which each is a literal (propositional variable or its negation), and
each disjunct is a clause. The problem is to find an assignment that satisfies . Given that
SAT is NP complete, systematic search methods can only solve problems of limited size. On the
other hand, relatively simple stochastic local search (SLS) methods have proved successful on a
wide range of larger and more challenging problems [10]. Furthermore, stochastic local search
(SLS) techniques are proven to be effective in solving hard satisfiability boolean problems.
However, their performance is still arguably poor when compared to systematic search
techniques. Therefore, and since the development of the first clause weighting dynamic local
search (DLS) algorithms for SAT, namely he Breakout heuristic [16], tries were made to enhance
the local search techniques in many different ways. Some researchers turned their focus on
studying the structure of the satisfiability problems (as in 1.1) to better understand the complexity

80 Computer Science & Information Technology (CS & IT)
of it and to come up with algorithms that could solve the problems in an optimal way. Other
researchers focused on investigating new heuristics that alter the search space based on some
information gathered during searching for a solution (as in 1.2). Thus, many heuristics,
enhancements and developments were introduced to improve SLS techniques performance in the
last three decades. As a result, a group of heuristics were introduced that could outperformed the
systematic search techniques, namely Dynamic Local Search (DLS). Recent DLS algorithms
depend on the use of weights to alter the search space. In other words, weights are used when
there are no moves that could decrease the search cost, to make it possible for the technique to
take unattractive moves which could increase the search cost temporarily.
1.1 Propositional satisfiability (SAT) complexity and hardness
It is proven that hard combinatorial SAT problems are the benchmarks that are used to test the
efficiency, accuracy and optimality of a given algorithm [12]. As easy problems could be solved
by any algorithm in a reasonable manner [21], which in turn does not reflect the real performance
of a solving techniques. Therefore, studies since almost three decades focused on studying the
hardness, complexity, and density of a countless number of SAT problems [6, 8, 15, 9]. Thus, a
large distribution of hard problems was produced. These hard problems are categorized into two
main divisions: 1) satisfiable and 2) un-satisfiable instances. Furthermore, in The International
SAT solver competition (http"//www.satcompetition.org/ ) there are three sub divisions of the
two main divisions: a) Industrial, b) Crafted, and c) Random instances. In another field of studies,
researchers not only investigated whether a problem is satisfiable or not, they also studied how
hard a problem is, as in [17, 21, 1].
1.2 Propositional satisfiability (SAT) dynamic solving techniques
Since the development of the Breakout heuristic [16], clause weighting dynamic local search
(DLS) algorithms for SAT have been intensively investigated, and continually improved [5, 7].
However, the performance of these algorithms remained inferior to their non-weighting
counterparts (e.g. [13]), until the more recent development of weight smoothing heuristics [24,
19, 11, 23]). Such algorithms now represent the state-of-the-art for stochastic local search (SLS)
methods on SAT problems. Interestingly, the most successful DLS algorithms (i.e. DLM [24],
SAPS [11], PAWS [23]), EWS [4], COVER [18] and recently CScore-SAT [3]) have converged
on the same underlying weighting strategy: increasing weights on false clauses in a local
minimum, then periodically reducing weights according to a problem specific parameter setting.
Except for COVER which updates the edge weights in every step of the search.
However, a key issue with DLS algorithms is that their performance depend mainly on the
efficiency of modifying the weights during the search, regardless of some other factors which
may play a crucial role in their performance when applied for solving large and hard SAT
problems such as Blocks World and Graph Coloring problems. For Instance, the size of
backbones [22]1
, the phase transition occurrence, and the density of a given problem.
Our study focuses on another factor and investigate its impact on the performance of DLS solving
techniques. The question addressed in the current paper is that what happens to the weights when
a clause and its neighboring clauses are satisfied?. For instance, if clause ci is connected to n
number of clauses (neighboring area of clause ci, as discussed in sub-section 2.2) and by

Computer Science & Information Technology (CS & IT) 81
assuming that clause ci became satisfied, by flipping one of its literals lim, where all its
neighboring clauses are satisfied too, should clause ci and its neighbors keep the weights?.
In the remainder of the paper we generally discuss the clause weighting most known algorithms
such as SAPS, PAWS and DLM and DDFW. Then we discuss elaborately on DDFW technique
as it is used for the purpose of the current study. Then, we show empirically the weights
behaviors and movements during the search via an intensive experimentation on a broad range of
benchmark SAT problems. Then we analyze the results and show the general outcome of the
experiments. Finally, we conclude our work and some guidelines for future work are given.
2. CLAUSE WEIGHTING FOR SAT
Clause weighting local search algorithms for SAT follow the basic procedure of repeatedly
flipping single literals that produce the greatest reduction in the sum of false clause weights.
Typically, all literals are randomly initialized, and all clauses are given a fixed initial weight. The
search then continues until no further cost reduction is possible, at which point the weight on all
unsatisfied clauses is increased, and the search is resumed, punctuated with periodic weight
reductions.
Existing clause weighting algorithms differ primarily in the schemes used to control the clause
weights, and in the definition of the points where weight should be adjusted. Multiplicative
methods, such as SAPS, generally adjust weights when no further improving moves are available
in the local neighborhood. This can be when all possible flips lead to a worse cost, or when no
flip will improve cost, but some flips will lead to equal cost solutions. As multiplicative real-
valued weights have much finer granularity, the presence of equal cost flips is much more
unlikely than for an additive approach (such as DLM or PAWS), where weight is adjusted in
integer units. This means that additive approaches frequently have the choice between adjusting
weight when no improving move is available, or taking an equal cost (flat) move.
Despite these differences, the three most well-known clause weighting algorithms (DLM [24],
SAPS [11] and PAWS [23]) share a similar structure in the way that weights are updated:2
Firstly, a point is reached where no further improvement in cost appears likely. The precise
definition of this point depends on the algorithm, with DLM expending the greatest effort in
searching plateau areas of equal cost moves, and SAPS expending the least by only accepting
cost improving moves. Then all three methods converge on increasing weights on the currently
false clauses (DLM and PAWS by adding one to each clause and SAPS by multiplying the clause
weight by a problem specific parameter α > 1). Each method continues this cycle of searching
and increasing weight, until, after a certain number of weight increases, clause weights are
reduced (DLM and PAWS by subtracting one from all clauses with weight > 1 and SAPS by
multiplying all clause weights by a problem specific parameter ρ< 1). SAPS is further
distinguished by reducing weights probabilistically (according to a parameter Psmooth), whereas
DLM and PAWS reduce weights after a fixed number of increases (again controlled by
parameter). PAWS is mainly distinguished from DLM in being less likely to take equal cost or
flat moves. DLM will take up to θ1 consecutive flat moves, unless all available flat moves have
______________
1
back in 2001, Slaney et. al. [22] studied the impact of backbones in optimization and approximation problems.
He concluded that in some optimization problems, backbones are correlated with the problem hardness. He also
suggested that heuristic methods when used to identify backbones may reduce problem difficulty.

already been used in the last θ2 moves. PAWS does away with these parameters, taking flat
moves with a fixed probability of 15%, otherwise it will increase weight.
2.1 Divide and Distribute Fixed Weights
DDFW introduces two ideas into the area of clause weighting algorithms for SAT. Firstly, it
evenly distributes a fixed quantity of weight across all clauses at the start of the search, and then
escapes local minima by transferring weight from satisfied to unsatisfied clauses. The other
existing state-of-the-art clause weighting algorithms have all divided the weighting process into
two distinct steps: i) increasing weights on false clauses in local minima and ii) decreasing or
normalizing weights on all clauses after a series of increases, so that weight growth does not
spiral out of control. DDFW combines this process into a single step of weight transfer, thereby
dispensing with the need to decide when to reduce or normalize weight. In this respect, DDFW is
similar to the predecessors of SAPS (SDF [19] and ESG [20]), which both adjust and normalize
the weight distribution in each local minimum. Because these methods adjust weight across all
clauses, they are considerably less efficient than SAPS, which normalizes weight after visiting a
series of local minima.3DDFW escapes the inefficiencies of SDF and ESG by only transferring
weights between pairs of clauses, rather than normalizing weight on all clauses. This transfer
involves selecting a single satisfied clause for each currently unsatisfied clause in a local
minimum, reducing the weight on the satisfied clause by an integer amount and adding that
amount to the weight on the unsatisfied clause. Hence DDFW retains the additive (integer)
weighting approach of DLM and PAWS, and combines this with an efficient method of weight
redistribution, i.e. one that keeps all weight reasonably normalized without repeatedly adjusting
weights on all clauses.
________
2
Additionally, a fourth clause weighting algorithm, GLSSAT [14], uses a similar weight update scheme,
additively increasing weights on the least weighted unsatisfied clauses and multiplicatively reducing
weights whenever the weight on any one clause exceeds a predefined threshold.
3
Increasing weight on false clauses in a local minimum is efficient because only a small proportion of the
total clauses will be false at any one time.

DDFW's weight transfer approach also bears similarities to the operations research sub-gradient
optimization techniques discussed in [20]. In these approaches, Lagrangian multipliers, analogous
to the clause weights used in SAT, are associated with problem constraints, and are adjusted in
local minima so that multipliers on unsatisfied constraints are increased and multipliers on
satisfied constraints are reduced. This symmetrical treatment of satisfied and unsatisfied
constraints is mirrored in DDFW, but not in the other SAT clause weighting approaches (which
increase weights and then adjust). However, DDFW differs from sub-gradient optimization in
that weight is only transferred between pairs of clauses and not across the board, meaning less
computation is required.
2.2 Exploiting Neighborhood Structure
Second and more original idea developed in DDFW, is the exploitation of neighborhood
relationships between clauses when deciding which pairs of clauses will exchange weight.
We term clause ci to be a neighbor of clause cj, if there exists at least one literal lim ϵ ci and a
second literal ljn ϵ cj such that lim = ljn as in Fig 1. Furthermore, we term ci to be a same sign
neighbor of cj if the sign of any lim ϵ ci is equal to the sign of any ljn ϵ cj where lim = ljn. From this it
follows that each literal lim ϵ ci will have a set of same sign neighboring clauses . Now, if ci
is false, this implies all literals lim ϵ ci evaluate to false. Hence flipping any lim will cause it to
become true in ci, and also to become true in all the same sign neighboring clauses of lim, i.e. it
will increase the number of true literals, thereby increasing the overall level of satisfaction for
those clauses. Conversely, lim has a corresponding set of opposite sign clauses that would be
damaged when lim is flipped.
The reasoning behind the DDFW neighborhood weighting heuristic proceeds as follows: if a
clause ci is false in a local minimum, it needs extra weight in order to encourage the search to
satisfy it. If we are to pick a neighboring clause cj that will donate weight to ci, we should pick
the clause that is most able to pay. Hence, the clause should firstly already be satisfied. Secondly,
it should be a same sign neighbor of ci, as when ci is eventually satisfied by flipping lim, this will
also raise the level of satisfaction of lim's same sign neighbors. However, taking weight from cj
only increases the chance that cj will be helped when ci is satisfied, i.e. not all literals in ci are
necessarily shared as same sign literals in cj , and a non-shared literal may be subsequently
flipped to satisfy ci. The third criteria is that the donating clause should also have the largest store
of weight within the set of satisfied same sign neighbors of ci.
The intuition behind the DDFW heuristic is that clauses that share same sign literals should form
alliances, because a flip that benefits one of these clauses will always benefit some other
member(s) of the group. Hence, clauses that are connected in this way will form groups that tend
towards keeping each other satisfied. However, these groups are not closed, as each clause will
have clauses within its own group that are connected by other literals to other groups. Weight is
therefore able to move between groups as necessary, rather than being uniformly smoothed (as in
existing methods).

3. EXPERIMENTAL EVALUATION AND ANALYSIS
Table 1. structural information about some of the the original DIMACS 2005 problem sets. the problem
were carefully selected taking to consideration their size, density, and connectivity.
As we stressed in section 2.1 and section 2.2, DDFW can exploit the neighboring structure of a
given clause ci and identify the weight alliances of ci. These weight alliances act as the source of
weights donors when a clause within the alliance need more weights. Also, they stabilize the
process of weight transfer as they can lead to keeping weights within each allies as long as no
weight transfer is needed, thus all the clauses within the neighborhood are satisfied. However,
this has further implications as it could lead the search to get into one of the following scenarios
that we discovered while investigating the weight transfer process: i) weights within a
neighborhood (Allie) could not be transferred to another sub area of the search space. ii) weights
of a specific neighborhood may keep circulating within their neighborhood. iii) if the level of
connectivity between any two neighboring allies is low, weight transfer may suffer from
stagnation, hence it can make the search process longer than it suppose to be.
In order to show the above three scenarios and their impact on the search process of any given
DLM technique, we first studied the general structure of some benchmark problems. Table 1
illustrates the structure of the benchmark problems that were carefully selected based on their

size, complexity and hardness. We attempted to reproduce a problem set similar to that used in
the random category of the SAT competition (as this is the domain where local search techniques
have dominated). To do this we selected the 50 satisfiable k3 problems from the SAT2004
competition benchmark. Secondly, we obtained the 10 SATLIB quasi-group existence problems
used in [2]. These problems are relevant because they exhibit a balance between randomness and
structure. Finally, we obtained the structured problem set used to originally evaluate SAPS [11].
These problems have been widely used to evaluate clause weighting algorithms (e.g. in [23]) and
contain a representative cross-section taken from the DIMACS and SATLIB libraries. In this set
we also included 4 of the well-known DIMACS 16-bit parity learning problems.
For each selected problem we firstly show the number of atoms (variables) of the problem, the
number of the clauses of the problem and the number of literals. Secondly, we show the
minimum and the maximum number of literals that form a clause within the problem structure.
Finally, we show the number of minimum sized clauses as will as the number of maximum sized
clauses. We designed our experiment to be as follows:
for each problem we ran DDFW 1000 run. Each run time out was set to 10,000,000 flips.
in each run, we recorded the change of weights in every 10,000 flips.
in every local minimum, we recorded whether DDFW heuristic selected a neighboring
satisfied clause from the weight allie of the false clause, or it picked a satisfied clause, to
be weight donor, randomly from outside the neighboring area of the false clause.
plots were made for each problem to illustrates the change of weights of the false clauses
from the starting point of the search process until the solution is found or it reaches a
time out. This is done for all the figures included in the paper.
Table 2 show the detailed results of the runs. For each problem we firstly show the success rate
(which reflect the percentage of whether a solution is found or not). Then we show the total
number of local minima that DDFW heuristics found before reaching a global optima. Then we
show the number of times the DDFW heuristic randomly picked a clause as a weight donor. Next
we show the number of times that DDFW heuristic picked a neighboring satisfied clause as a
weight donor from the false clause allies. Finally we show the the average number of times
DDFW heuristic deterministically picked a clause as a weight donor.
In order to show the weights transfer and movements during the search we plotted the false
clauses and their weights changes and the number of neighboring clauses of each false clause.
This to show the relationship between the change of weights and the number of neighboring
clauses of a false clause. Out of all problem sets we discuss four problems as they are of great
importance to this work, namely, the Uniform Random 3SAT, the Parity problem, the Blocks
World and the Graph Coloring problem because they explicitly show the previously mentioned
three scenarios. These four problem sets are of different sizes and level of hardness. The Uniform
Random 3SAT (uf100 and uf250) is the smallest set that has 100 variables and 430 clauses,
where the Uniform Random 3SAT 250 has 250 variables and 1065 clauses. Fig 2 show the results
for both problems. We can see that both problems were easy to solve. Also, the weight movement
was smooth as their was enough neighboring clauses to donate weights to a false clause and more
importantly when a sub area become satisfied, the connectivity between clauses allow the transfer

of weights easily (no weight stagnations occur). This is also true with the second problem, The
Parity problem even that the level of hardness of the Parity 16 and the Parity 32 is higher than the
Uniform Random 3SAT, as in Fig 3. What was experimentally interesting is the Graph Coloring
problem (both the g125.17 and the g125.18 problems as in Fig 4, where g125.17 has 2125
variables 66272 clauses and the g125.18 has 2250 variables and 70163 clauses ) where firstly,
DDFW could not reach 100% success rate on the g125.17. Secondly, the figure show a clear gap
between the movement of the weights. Which means the occurrence of weights stagnation. Thus,
the connectivity between the clauses is either very low which make transferring weights among
the clauses is limited to the clauses that are directly connected, or the connectivity of the clauses
is very high which means a larger number of neighbors that could keep the weights for longer
time and prevent other false clauses some where else in the search space from using them. Finally
the Blocks World, Fig 5, which has 6325 variables and 131973 clauses., which has a similar
weight transfer behaviors as the parity problem and the uniform random 3SAT problem with the
exception of the level of hardness as the block world problem was harder to solve and the weights
were moving more often during the search space.
Table 2. The table show the number of successful tries made by DDFW, the number of local minima faced
the search, the number of random weights distribution during the search, the number of deterministic
weight distribution and the average number of deterministc weight distribution. Weights behaviors and
movements during the search.

Fig. 2. false clauses and their weights during the search, the uf100 problem top and the uf250 bottom
Fig. 3. false clauses and their weights during the search, the par16 top and par32 bottom

Fig. 4. false clauses and their weights during the search, the g125-17 top and the g125-18 bottom
Fig. 5. false clauses and their weights during the search, the bw-d.large
4. CONCLUSION
As a conclusion, DLS weighting techniques performance could suffer from weight stagnation that
leads to the slowness of a chosen techniques. Our experiments show that these weight stagnations
are not general to all problems but rather they are a problem specific characteristic. We have
looked into each problem characteristic such as its hardness, complexity, and density. As a result,
each problem characteristics may contribute to the occurrence of weight stagnations in some
stages during the search. The DDFW algorithm is a relatively simple application of neighborhood
weighting, and further experiments (not reported here) indicate more complex heuristics can be
more effective on individual problems. In particular, we have looked at adjusting the amount of
weight that is redistributed and allowing DDFW to randomly pick donor clauses according to a
noise parameter. However, we have yet to discover a general neighborhood heuristic as effective

as DDFW over the range of problems considered. In future work we consider it will be promising
to extend a DDFW-like approach to handle weight stagnations via adjusting weights in stagnated
alliances regardless of whether they are satisfied or not. This could be done by improving the
exploitation of neighboring areas.
ACKNOWLEDGEMENT
The authors would like to acknowledge the financial support of the Scientific Research
Committee at Petra University. Also we would like to thank all faculty members of the
Information Technology faculty who contributed directly and indirectly to this work.
REFERENCES
[1] D. Achlioptas, C. Gomes, H. A. Kautz, and B. Selman, Generating satisfiable instances," in
Proceedings of 17th AAAI, 2000, pp. 256{261.
[2] Anbulagan, D. Pham, J. Slaney, and A. Sattar, Old resolution meets modern SLS," in Proceedings of
20th AAAI, 2005, pp. 354{359.
[3] S. Cai, C. Luo, and K. Su, Scoring functions based on second level score for k-sat with long clauses,"
J. Artif. Intell. Res. (JAIR), vol. 51, pp. 413{441, 2014.
[4] S. Cai, K. Su, and Q. Chen, EWLS: A new local search for minimum vertex cover," in Proceedings
of the Twenty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2010, Atlanta, Georgia,
USA, July 11-15, 2010, 2010.
[5] B. Cha and K. Iwama, Adding new clauses for faster local search," in Proceedings of 13th AAAI,
1996, pp. 332{337.
[6] S. A. Cook, The complexity of theorem-proving procedures," in Proceedings of the Third Annual
ACM Symposium on Theory of Computing, ser. STOC'71. New York, NY, USA: ACM, 1971, pp.
151{158. [Online]. Available: http://doi.acm.org/10.1145/800157.805047
[7] J. Frank, Learning short-term clause weights for GSAT," in Proceedings of 15th IJCAI, 1997, pp.
384{389.
[8] M. R. Garey and D. S. Johnson, Computers and Intractability; A Guide to the Theory of NP-
Completeness. New York, NY, USA: W. H. Freeman & Co., 1990.
[9] I. P. Gent and T. Walsh, The hardest random SAT problems," in KI-94: Advances in Artificial
Intelligence, 18th Annual German Conference on Artificial Intelligence, Saarbrucken, Germany,
September 18-23, 1994, Proceedings, 1994, pp. 355{366.
[10] H. Hoos and T. Stulze, Stochastic Local Search. Cambridge, Massachusetts: Morgan Kaufmann,
2005.
[11] F. Hutter, D. Tompkins, and H. Hoos, Scaling and Probabilistic Smoothing: Efficient dynamic local
search for SAT," in Proceedings of 8th CP, 2002, pp. 233{248.
[12] M. Jarvisalo, D. Le Berre, O. Roussel, and L. Simon, The international SAT solver competitions," AI
Magazine, vol. 33, no. 1, pp. 89{92, 2012.

[13] D. McAllester, B. Selman, and H. Kautz, Evidence for invariants in local search," in Proceedings of
14th AAAI, 1997, pp. 321{326.
[14] P. Mills and E. Tsang, Guided local search applied to the satisfiability (SAT) problem," in
Proceedings of 15th ASOR, 1999, pp. 872{883.
[15] D. Mitchell, B. Selman, and H. Levesque, Hard and easy distributions of sat problems," in
ghiuyguyigy, 1992, pp. 459{465.
[16] P. Morris, The Breakout method for escaping from local minima," in Proceedings of 11th AAAI,
1993, pp. 40{45.
[17] P. Prosser, Binary constraint satisfaction problems: Some are harder than others," in ECAI.
PITMAN, 1994, pp. 95{95.
[18] S. Richter, M. Helmert, and C. Gretton, A stochastic local search approach to vertex cover," in KI
2007: Advances in Artificial Intelligence, 30th Annual German Conference on AI, KI 2007,
Osnabruck, Germany, September 10-13, 2007, Proceedings, 2007, pp. 412{426.
[19] D. Schuurmans and F. Southey, Local search characteristics of incomplete SAT procedures," in
Proceedings of 10th AAAI, 2000, pp. 297{302.
[20] D. Schuurmans, F. Southey, and R. Holte, The exponentiated subgradient algorithm for heuristic
boolean programming," in Proceedings of 17th IJCAI, 2001, pp. 334{341.
[21] B. Selman, D. Mitchell, and H. Levesque, Generating hard satisfiability problems," Artificial
Intelligence, vol. 81, pp. 17{29, 1996.
[22] J. Slaney and T. Walsh, Backbones in optimization and approximation," in Proceedings of the 17th
International Joint Conference on Artificial Intelligence - Volume1, ser. IJCAI'01, 2001, pp. 254{259.
[23] J. Thornton, D. N. Pham, S. Bain, and V. Ferreira Jr., Additive versus multiplicative clause
weighting for SAT," in Proceedings of 19th AAAI, 2004, pp. 191{196.
[24] Z. Wu and B. Wah, An efficient global-search strategy in discrete Lagrangian methods for solving
hard satisfiability problems," in Proceedings of 17th AAAI, 2000, pp. 310{315

WEIGHTS STAGNATION IN DYNAMIC LOCAL SEARCH FOR SAT

More Related Content

Similar to WEIGHTS STAGNATION IN DYNAMIC LOCAL SEARCH FOR SAT (20)

More from cscpconf (20)

Recently uploaded (20)

WEIGHTS STAGNATION IN DYNAMIC LOCAL SEARCH FOR SAT